Educational and Psychological Measurement

Educational and Psychological Measurement http://epm.sagepub.com/

Investigating Validity Evidence for the Experiences in Close Relationships-Revised Questionnaire Amanda J. Fairchild and Sara J. Finney Educational and Psychological Measurement 2006 66: 116 DOI: 10.1177/0013164405278564 The online version of this article can be found at: http://epm.sagepub.com/content/66/1/116

Published by: http://www.sagepublications.com

Additional services and information for Educational and Psychological Measurement can be found at: Email Alerts: http://epm.sagepub.com/cgi/alerts Subscriptions: http://epm.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav Citations: http://epm.sagepub.com/content/66/1/116.refs.html

Downloaded from epm.sagepub.com at UNIV OF SOUTH CAROLINA on September 20, 2010

Educational 10.1177/0013164405278564 Fairchild, Finney and Psychological / Validity Evidence Measurement for the ECR-R Questionnaire

Investigating Validity Evidence for the Experiences in Close Relationships– Revised Questionnaire

Educational and Psychological Measurement Volume 66 Number 1 February 2006 116-135 © 2006 Sage Publications 10.1177/0013164405278564 http://epm.sagepub.com hosted at http://online.sagepub.com

Amanda J. Fairchild Arizona State University, Tempe

Sara J. Finney James Madison University, Harrisonburg, Virginia

The current study gathered internal structural validity and external criterion validity evidence for the Experiences in Close Relationships–Revised Questionnaire (ECR-R) scores. Specifically, confirmatory factor analysis of the data provided general support for the hypothesized two-factor model, and hypothesized relationships with external criteria 2 were substantiated. However, minor model misfit and low communalities (R ) suggested that some items may represent extraneous constructs. Further avenues of study regarding the functioning of the instrument are provided. Keywords: attachment; close relationships; construct validity; ECR-R

T

he purpose of the present study is to investigate reliability and validity evidence for a quantitative self-report measure of adult romantic attachment. Investigations of this kind are particularly valuable, as adult attachment style has been shown to have consequences for behavior in intimate relationships (e.g., Feeney, Noller, & Hanrahan, 1994; Morrison, Goodlin-Jones, & Urquiza, 1997). Given that research in the field hinges on the legitimacy of the measurement tools themselves, studying such instruments may only further inform the domain.

Why Study Attachment? Researchers have previously established a connection between attachment and satisfaction in interpersonal relationships (Collins & Read, 1990; Feeney et al., 1994;

Authors’Note: Correspondence concerning this article should be addressed to Amanda J. Fairchild, Department of Psychology, Arizona State University, PO Box 871104, Tempe, AZ 85287-1104; e-mail: amanda [email protected].

116


Fairchild, Finney / Validity Evidence for the ECR-R Questionnaire 117

Hazan & Shaver, 1987; Morrison et al., 1997). In particular, Feeney et al. (1994) ascertained that individual “anxiety about attachment issues is the driving force behind a range of negative and destructive patterns of communication” that develop within the dyad (p. 284). Other sources (Bartholomew & Horowitz, 1991; Collins, 1996; Hazen & Shaver, 1987) support this claim, emphasizing that securely attached individuals express more confidence in their relationships with others, have a greater capacity for trust, and effectively support their intimate partners; whereas insecure individuals consistently display more anxiety or avoidance of personal relationships, higher degrees of jealousy, and greater levels of dissatisfaction in close relationships. Unfortunately, the field of adult attachment theory is lacking in terms of a unified quantitative measurement methodology. Additionally, those instruments that do exist have limited reliability and validity evidence for their scores (Bartholomew & Horowitz, 1991; Becker, Billings, Eveleth, & Gilbert, 1997; Collins & Read, 1990; Griffin & Bartholomew, 1994; Hazan & Shaver, 1987; Simpson, 1990). Legitimate research in the domain requires meaningful interpretation of scores from measures that adequately represent the attachment construct within the romantic relationship (i.e., not developmental attachment measures), and recent efforts in the discipline have attempted to approach this standard. The current investigation will consider both internal and external validity evidence for scores from one such instrument, the Experiences in Close Relationships–Revised (ECR-R; Fraley, Waller, & Brennan, 2000) scale. However, before considering the scale in question, it is useful to discuss the theory behind its creation and the scale that preceded it.

Developmental Attachment Theory Bowlby’s work in the late sixties and early seventies developed the original framework for attachment theory. His work conveyed that the quality of early attachment experiences had lasting effects on child development (1973, 1977, 1982). Bowlby identified three components of the attachment relationship, which acted as the “how” and “why” behind its functioning: (a) proximity maintenance with the caregiver; (b) caregiver safe haven for infants; and (c) a secure base, from which the infant could explore its environment and engage in activities unrelated to attachment (Bowlby, 1979, 1982). Ainsworth, Blehar, Waters, and Wall (1978) complemented Bowlby’s research by delineating individual differences that occurred in the development of these early attachment relationships. Their observations yielded the classification of three different attachment styles: secure, anxious, and avoidant. These categories were conceptualized within a two-dimensional space, with anxiety and avoidance as the two dimensions (see Figure 1). Importantly, the three categories were later supplemented by the addition of a fearful attachment style orientation (high anxiety, high avoidance), producing a four-category classification system (Bartholomew & Horowitz, 1991).


118 Educational and Psychological Measurement

Figure 1 Ainsworth, Blehar, Waters, and Wall’s (1978) Individual Differences of Attachment

ANXIETY High Anxiety

Anxious

Low Avoidance

High Avoidance

AVOIDANCE Avoidant

Secure

Low Anxiety

Adult Attachment Theory These attachment styles, which develop in infancy, are hypothesized to continue affecting an individual’s intimate and social life over the life course. The internal working models that emerge in childhood provide the means by which individuals regard their personal relationships and develop expectations about future relationships in adulthood (Hazan & Shaver, 1994; Morrison et al., 1997). Specifically, an individual mentally establishes whether partners will respond to his or her needs for support and attention and whether he or she feels worthy of being the recipient of such support. Ultimately, this cognitive framework influences how the individual acts with close others throughout the life course (Ainsworth 1989; Bowlby 1973, 1980, 1982).

Measurement in the Adult Attachment Literature Although the application of developmental attachment theory to the close relationship literature incited much new research and scale development in the field, these endeavors lacked convergence (Griffin & Bartholomew, 1994). Researchers’ attempts



at creating a body of empirical work not only left areas uninvestigated but also yielded significant overlap in the resulting instrumentation. In response to this, Brennan, Clark, and Shaver (1998) developed the ECR, seeking to create a measure that would serve as the domain standard. An exhaustive search of the adult attachment literature led to retaining 323 items and 60 separate subscales for the instrument after redundant content was removed. This effort superseded any other romantic attachment inventory to date in that it was the most integrative measure of its kind. Brennan, Clark, et al. (1998) constructed their scale based on Ainsworth et al.’s (1978) two-dimensional theory model of attachment. However, like Bartholomew and Horowitz (1991), Brennan, Clark, et al. believed that there were four possible attachment categories associated with the anxiety and avoidance dimensions. Importantly, the dimensional nature of the ECR scores allowed for differentiation among attachment styles, while preventing the loss of information that typically accompanies categorical techniques (Fraley & Waller, 1998; Griffin & Bartholomew, 1994). More specifically, gathering data concerning respondents’ level of anxiety and avoidance provided more information than simply asking respondents to check which attachment style best represented them. Principal components analysis yielded two components formed from the ECR items, which Brennan, Clark, et al. (1998) interpreted as anxiety and avoidance. The two-component solution that they retained accounted for almost 63% of the variance in the items. In addition, Cronbach’s coefficient alpha for scores from the Anxiety (α = .91) and Avoidance (α = .94) subscales reflected high internal consistency. To create a short measure of adult attachment, Brennan, Clark, et al. (1998) retained the 18 items with the highest absolute structure weights for each component; this yielded the continuous 36-item measure called the ECR. To discuss their results in terms of the categorical and prototypical scales used by their predecessors (e.g., Bartholomew & Horowitz, 1991; Becker et al., 1997; Collins & Read, 1990; Griffin & Bartholomew, 1994; Hazan & Shaver, 1987; Simpson, 1990), Brennan, Clark, et al. also used a hierarchical clustering procedure to place people into one of four groups, where assignment to a given attachment type was based upon the subject’s scores on the two attachment components. Though the cluster analysis essentially yielded the same formation of “groups” or “categories” as in earlier scales, this approach allowed researchers to gather information about the subject’s degree of endorsement of anxiety and avoidance, rather than simply categorizing subjects into one of four absolute attachment types. Validity evidence gathered for scores on the new measure supported hypothesized relationships with two theoretically related constructs: touch and sexuality. Moreover, the ECR was found to predict scores on these constructs better than other adult attachment measures. Specifically, the new measure accounted for almost 3 times the variance in the touch and sexuality scores than the earlier Bartholomew and Horowitz (1991) instrument. Fraley et al. (2000) subsequently revised the ECR using item response theory (IRT). The test information curves for the two represented dimensions were much higher than other studied attachment scales (e.g., Collins & Read, 1990; Griffin &



Bartholomew, 1994; Simpson, 1990), and analysis generally indicated a good degree of measurement precision across different levels of the construct. However, the ECR, as well as the other scales studied, was less precise at measuring people with secure attachment styles (i.e., low anxiety and low avoidance). In an effort to improve this measurement precision, Fraley et al. reanalyzed the original 323-item data set from Brennan, Clark, et al. (1998) and retained items based on their discrimination values alone. This resulted in the ECR-R, where the new 18-item Anxiety subscale retained 13 out of the original 18 items, and the new 18-item Avoidance subscale retained 7 out of the original 18 items. Subsequent IRT analysis on the revised 36-item set indicated substantial improvement in the scale’s measurement precision of the dimensions, while also demonstrating good individual item functioning. Problematically, though, little research has focused on collecting construct validity and reliability evidence for scores from the new measure. These investigations are especially important as the ECR-R subscales differ considerably from their predecessor.

Steps in Instrument Development Benson (1998) described a three-step approach to developing a strong program for construct validity, which includes (a) a substantive component, (b) a structural component, and (c) an external component. The substantive component of the program involves theoretically and empirically defining the domain of interest, so that potential items of a construct are adequately represented in measurable ways. It can be argued that Brennan, Clark, et al. (1998) thoroughly researched the theoretical domain of attachment in adult romantic relationships when they developed the 36-item ECR from 323 nonredundant items. Because the creation of the ECR-R was derived from the same item pool, one can say that the ECR-R substantively borrowed from the ECR instrument. Thus, the stage was set for internal domain studies that constitute the structural component of the program, which focuses on relating items both to one another and to the construct of interest to clarify interrelationships. Fairchild, Pastor, and Brennan (2003) conducted an initial principal-axis factor analysis on the ECR-R. The results of their study supported Brennan, Clark, et al.’s original two-factor conceptualization of attachment, providing some internal domain support for the construct validity of the scores. However, to date the external component of a validity investigation has not been conducted on scores from the revised measure. This leaves the question of how the ECR-R scores relate to other constructs. Therefore, the present study aimed to further investigate validity evidence for the ECR-R scores. Specifically, this investigation examined the three following three research questions: Research Question 1: Does a two-factor model of adult attachment underlie the responses to the ECR-R items? Research Question 2: Will the internal consistency estimates of the scores for each subscale approximate the reliabilities found for scores from the subscales of the original ECR investigation (α = .91-.94)? Research Question 3: Will correlations between scores from the ECR-R and theoretically related constructs provide evidence of construct validity for the scale scores?



A number of different criteria were employed to investigate the third research question, namely, touch, loneliness, perceived social support, and worry. Little research has been conducted on the relationship between touch and adult attachment, as both constructs are relatively new research domains. However, Brennan, Clark, et al. (1998) found that scores from the ECR accounted for large amounts of variance in scores from the Brennan, Wu, and Loev (1998) touch measure. Beyond this evidence, touch may be inextricably linked to the formation of developmental attachment bonds, as the entire premise of the attachment system focuses on sustaining proximity with one’s caregiver. Because these developmental experiences are purported to influence relationships into adulthood, it follows that touch may play an integral role in adult attachment. Therefore, although previous empirical research has not established an explicit directional relationship between avoidance and touch, tentative predictions are still reasonable. Specifically, it is hypothesized that avoidance, as assessed through the ECR-R Avoidance subscale, will have moderate (r = .25-.40) to strong (r = .60.80), positive correlation with touch aversion as measured by the Touch Scale (Brennan, Wu, et al., 1998). Additionally, it is hypothesized that avoidance will have a moderate (r = .25-.40) to strong (r = .60-.80), negative correlations with using touch as a means to express affectionate proximity, desiring more touch, and using touch to attain a safe haven, as measured by the same instrument. The relationship between loneliness and attachment style may also be best assessed through an examination of developmental attachment theory. Through early attachment experiences, individuals come to develop either positive or negative views about themselves and others in personal relationships (Bowlby, 1973). Because negative self-views involve believing that one is unworthy of being cared for, these individuals may be less adept at seeking out and sustaining relationships. Therefore, negative selfmodels may have the potential to preclude the development of relationships (Bowlby, 1977). It follows, then, that insecurely attached individuals, particularly those with high degrees of anxiety, may experience greater degrees of loneliness than those individuals who are more secure and less anxious. In fact, research indicates that those attachment styles with high degrees of anxiety have been associated with greater levels of loneliness (Man & Hamid, 1998). Therefore, it is hypothesized that anxiety, as measured by the ECR-R Anxiety subscale, will have a moderate (r = .25-.40) to strong (r = .60-.80), positive correlation with loneliness as assessed through the UCLA Loneliness Scale (Russell, 1996). Additionally, it is hypothesized that avoidance will also have a moderate (r = .25-.40) positive correlation with loneliness. Previous research has also offered evidence in support of the relationship between attachment and social support (Bartholomew & Horowitz, 1991; Davis, Morris, & Kraus, 1998; Florian, Mikulincer, & Bucholtz, 1995). A large component of Bowlby’s (1982) developmental attachment theory involves individuals’ seeking proximity to a caregiver in times of need and distress to promote safety and well-being. Therefore, if an individual experiences a satisfying attachment relationship (i.e., experiences low anxiety and low avoidance), he or she is more likely to seek proximity and to have high degrees of perceived social support. In contrast, those individuals whose attachment relationships are more characterized by high anxiety and avoidance should be less



likely to seek proximity and therefore less likely to perceive social support. Empirical research has supported this claim, submitting that individuals who were characterized as more avoidant in their attachment orientation sought out less social support than those who were less avoidant (Brennan & Shaver, 1995; Davis et al., 1998). Research has also shown that individuals who report higher degrees of anxiety report lesser degrees of social support as well (Hawkins, 1995). As such, it is hypothesized that there will be a moderate (r = .25-.40) to strong (r = .60-.80), negative correlation between the Avoidance subscale of the ECR-R and perceived social support as measured by the Social Provisions Scale (SPS; Cutrona & Russell, 1987). It is also hypothesized that there will be a moderate (r = .25-.40) to strong (r = .60-.80), negative correlation between anxiety and perceived social support. Finally, the notion that worry should relate to anxiety is evident in the definition of anxiety itself, or “a state of uneasiness and distress about future uncertainties; apprehension; worry” (Berube et al., 1982). Moreover, worry is listed as a diagnostic criterion for Generalized Anxiety Disorder (GAD) as characterized by the Diagnostic and Statistical Manual of Mental Disorders, 4th edition (DSM-IV; American Psychiatric Association, 1994). Gana, Bettina, and Canouet (2001) have gone further to elucidate a causal relationship between worry and anxiety, such that worry exerts a statistically significant, positive effect on anxiety. Accordingly, a moderate (r = .25-.40) to strong (r = .60-.80), positive correlation between anxiety, as measured by the ECR-R Anxiety subscale (Brennan, Clark, et al., 1998), and worry, as measured by the Penn State Worry Questionnaire (PSWQ; Meyer, Miller, Metzger, & Borkovec, 1990), is hypothesized.

Method Participants and Procedure Four hundred twenty-nine students from an undergraduate psychological subject pool at a midsized, southeastern university participated in the study. The study sample was composed of 61% females. The majority of participants were either 18 (n = 149) or 19 (n = 147) years old. Of the total 429 participants, 93.9% had been in a romantic relationship at least once before. Due to the subject matter of the questionnaire, those individuals who reported never having been in a romantic relationship were subsequently removed from analysis. Listwise deletion of missing data on this reduced sample indicated that 397 participants had a complete set of responses to the ECR-R, whereas 370 participants had a complete set of responses to every measure. Prior to test administration, students were read a standardized script and advised that their participation in the study was voluntary. After signing an informed consent form, all students received a manila envelope containing the test battery. Participants were told that they would complete each instrument one at a time and that no one person would be allowed to begin the next measure until everyone had finished. The instructions for each instrument were read aloud by the experimenter prior to the administration of each. This method attempted to slow down response rates, in an



effort to promote thoughtful responding to each instrument. Although the ECR-R was the first instrument administered across all experimental sessions, the rest of the instruments were counterbalanced across each session. It was determined that the ECR-R should be presented first across all sessions, as it was the primary scale of interest. Furthermore, it was anticipated that counterbalancing all other instruments in the test battery would circumvent any order or fatigue effects that may have otherwise occurred. The instruments took approximately 25 minutes to complete, and participants were fully debriefed upon completion of the study.

Instrumentation Experiences in Close Relationships–Revised (ECR-R; Fraley et al., 2000). The ECR-R is a 36-item measure used to assess adult romantic attachment. Participants were instructed to indicate how they generally experience relationships. Respondents used a 7-point Likert-type scale ranging from 1 (disagree strongly) to 7 (agree strongly), such that higher scores were associated with a higher endorsement of the construct. As discussed above, the scale consisted of two 18-item subscales: Anxiety and Avoidance. An example of an item representing avoidance is “I find it difficult to allow myself to depend on romantic partners.” An example of an item representing anxiety is, “I often worry that my romantic partner doesn’t really love me.” Touch Scale (Brennan, Wu, et al., 1998). The Touch Scale is a 51-item instrument that represents seven touch dimensions within the context of romantic relationships: (a) desire for touch, (b) affectionate proximity, (c) sexual touch, (d) touch aversion, (e) discomfort with public touch, (f) coercive control, and (g) safe-haven touch. Responses to the items were measured using a 7-point scale that ranged from 1 (not at all like me) to 7 (very much like me), such that higher scores represented higher endorsement of the latent construct. Brennan, Wu, et al. (1998) derived the item pool from all available touch inventories to create an integrative measure representing various dimensions of touch in romantic relationships. Construct validity evidence for the scores was provided by correlations between scores from the measure and scores from related constructs. Estimates of internal consistency for the scores ranged from α = .71 to α = .86. Replicating Brennan, Clark, et al.’s (1998) validity investigation of the original ECR, the current investigation only considered four of the seven subscales: (a) using touch as a means to express affectionate proximity, (b) desire for touch, (c) aversion to touch, and (d) using touch to attain a safe-haven. In the current study, Cronbach’s coefficient alpha estimates for the four subscale scores were (a) α = .89 for the Affectionate Proximity subscale, (b) α = .87 for the Desire for Touch subscale, (c) α = .88 for the Aversion to Touch subscale, and (d) α = .74 for the Safe-Haven Touch subscale. An example of an item representing affectionate proximity is “I usually hug my partner to show how happy I am to see him or her.” An example of an item representing the desire for more touch is “Sometimes I am not very happy with the level of touch in my relationship.” An example of an item representing touch aversion is “I often have to remind my partner to stop touching me”; and finally, an example of an



item representing safe-haven touch is “When I am facing a difficult situation, I like being touched by my partner.” UCLA Loneliness Scale–Version Three (Russell, 1996). The UCLA Loneliness Scale–Version Three is a 20-item instrument that represents a unidimensional construct of loneliness. Loneliness was measured using a 4-point scale that ranged from 1 (never) to 4 (always), such that higher scores were associated with greater feelings of loneliness. An example of a question from the scale is “How often do you feel isolated from others?” Construct validity evidence for the scores was provided in the form of correlations between scores from the measure and scores from other loneliness instruments. In addition, hypothesized relationships between both social support and life satisfaction were supported. Internal consistency estimates for the scores with reference to college students was α = .92. The test-retest reliability estimate for the scores over a one-year period was substantial, r = .73. In the current study, α = .93. The Social Provisions Scale (SPS; Cutrona & Russell, 1987). The SPS is a unidimensional 24-item scale used to measure a subject’s perceived degree of social support. A 4-point scale that ranges from 1 (strongly disagree) to 4 (strongly agree) is used to respond to each item, such that higher scores are associated with higher levels of perceived social support. An example of an item on the scale is “There is someone I could talk to about important decisions in my life.” Convergent validity of the scores was assessed by investigating correlations between scores from the measure and scores from a number of other social support measures in the literature. Discriminant validity coefficients were derived from differentiating between the scores on the measure and scores from a social desirability instrument, whereas divergent validity coefficients were derived from relationships with depression, introversion-extraversion, neuroticism, and number of stressful events measures. Internal consistency estimates for the scores during development were α = .92. In the current study, α = .84. The Penn State Worry Questionnaire (PSWQ; Meyer et al., 1990). The PSWQ is designed to measure levels of worry in both clinical and nonclinical samples. A 5point scale that ranges from 1 (not at all typical of me) to 5 (very typical of me) is used to respond to each item, such that higher scores were associated with higher levels of worry. An example of an item on the scale is “I notice that I have been worrying about things.” The development of the 16-item PSWQ resulted from a series of eight different studies conducted on the instrument. Convergent validity of the scores was supported by high correlations between scores from the PSWQ and scores from the trait subscale of the Spielberger State-Trait Anxiety Inventory, as well as other worryrelated constructs, such as perfectionism. Divergent validity for the scores was supported by appropriate negative correlations between the PSWQ and several different psychological constructs (e.g., self-esteem). Internal consistency estimates for the scores ranged from α = .93 to α = .95. Test-retest reliability for the scores was r = .92. In the current study, α = .95.



Results Descriptive Statistics Interitem correlations for the ECR-R are available upon request from the first author. Means, standard deviations, skewness, and kurtosis for the ECR-R items are presented in Table 1. Standard deviations of the items ranged from 1.26 to 1.86, indicating variability in the item responses. Furthermore, both univariate skewness and kurtosis indices were within acceptable ranges indicating adequate univariate normality. That is, neither skew nor kurtosis exceeded recommended cutoffs, |3.00| and |8.00|, respectively (Kline, 1998). Given the fairly normal distribution of scores, maximum likelihood (ML) estimation was employed. ML was considered desirable as this estimator has been shown to yield the most asymptotically unbiased, efficient, and consistent parameter estimates of any estimation method under conditions of normality (West, Finch, & Curran, 1995).

Confirmatory Factor Analysis (CFA): Model Fit CFA provides the opportunity to test various a priori specified factor structures that may underlie responses to a given set of variables. The models considered in the present analysis were (a) a two-factor model representing anxiety and avoidance as two distinct, yet correlated, factors of the adult attachment construct; and (b) a nested onefactor model that considered a unified adult attachment factor. A chi-square difference test (∆χ2) was employed to compare the fit of the two nested models with respect to one another. 2 In addition to examining the ∆χ , various fit indices were also examined to assess model fit. Specifically, both absolute and incremental fit indices were utilized. Absolute indices simply consider how well the model accounts for observed covariances in the data (Hu & Bentler, 1995), whereas incremental fit indices consider the improvement in fit of the proposed model over an independence baseline model. Based upon previous work examining the sensitivity of various fit indices (Hu & Bentler, 1998, 1999), the comparative fit index (CFI), the root mean square error of approximation (RMSEA), and the standardized root mean square residual (SRMR) were used to assess model fit. CFA analyses were conducted with LISREL 8.54 (Jöreskog & Sörbom, 1993), and the variance/covariance matrix of the scores was used as input in analysis. The various fit indices for the one- and two-factor models are presented in Table 2. Results indicate that none of the indices associated with the one-factor model met, or approached, recommended standards (i.e., CFI ≥ .95, RMSEA ≤ .06, SRMR ≤ .08; Hu & Bentler, 1999). Furthermore, the ∆χ2 between the one- and two-factor models yielded statisti2 cally significant results, χ (1) = 1,356.59, p < .001, suggesting that the one-factor model fit statistically worse than the two-factor model. Also, the various fit indices associated with the two-factor model indicated superior fit. Specifically, the SRMR



Table 1 Item Means, Standard Deviations, Skewness, and Kurtosis for the Experiences in Close Relationships–Revised (ECR-R) Items Item (1) I prefer not to show a partner how I feel deep down. (2) I often worry that my partner will not want to stay with me. (3) I feel comfortable sharing my private thoughts and feelings with my partner. (R) (4) I’m afraid that I will lose my partner’s love. (5) I find it difficult to allow myself to depend on romantic partners. (6) I often worry that my romantic partner doesn’t really love me. (7) I am very comfortable being close to romantic partners. (R) (8) I worry that romantic partners won’t care about me as much as I care about them. (9) I don’t feel comfortable opening up to romantic partners. (10) I often wish that my partner’s feelings for me were as strong as my feelings for him or her. (11) I prefer not to be too close to romantic partners. (12) I worry a lot about my relationships. (13) I get uncomfortable when a romantic partner wants to be very close. (14) When my partner is out of sight, I worry that he or she might become interested in someone else. (15) I find it relatively easy to get close to my partner. (R) (16) When I show my feelings for romantic partners, I’m afraid they will not feel the same about me. (17) It’s not difficult for me to get close to my partner. (R) (18) I rarely worry about my partner leaving me. (R) (19) I usually discuss my problems and concerns with my partner. (R) (20) My romantic partner makes me doubt myself. (21) It helps to turn to my romantic partner in times of need. (R) (22) I do not often worry about being abandoned. (R) (23) I tell my partner just about everything. (R) (24) I find that my partner(s) don’t want to get as close as I would like. (25) I talk things over with my partner. (R) (26) Sometimes romantic partners change their feelings about me for no apparent reason. (27) I am nervous when partners get too close to me. (28) My desire to be very close sometimes scares people away. (29) I feel comfortable depending on romantic partners. (R) (30) I’m afraid that once a romantic partner gets to know me, he or she won’t like who I really am. (31) I find it easy to depend on romantic partners. (R) (32) It makes me mad that I don’t get the affection and support I need from my partner.

M

SD

Skewness

Kurtosis

2.73 3.55

1.56 1.60

0.783 –0.021

–0.261 –1.075

2.64 3.32

1.52 1.60

1.015 0.189

0.317 –0.964

3.44

1.69

0.215

–0.878

2.90 2.61

1.58 1.51

0.496 0.934

–0.799 0.185

3.79 2.81

1.77 1.57

–0.020 0.722

–1.115 –0.325

3.72 2.54 3.69

1.78 1.56 1.81

0.043 0.923 0.155

–0.907 0.024 –1.118

2.61

1.58

0.881

–0.152

3.47 2.85

1.78 1.46

0.269 0.626

–0.999 –0.178

3.53 3.02 3.50

1.67 1.60 1.59

0.139 0.599 0.092

–1.004 –0.478 –1.069

2.56 2.20 2.27 3.13 2.88

1.36 1.45 1.37 1.76 1.70

0.774 1.285 1.355 0.452 0.700

0.061 0.963 1.827 –0.916 –0.496

2.87 2.39

1.50 1.26

0.494 0.913

–0.592 0.728

3.02 2.74 2.63 3.31

1.72 1.63 1.64 1.61

0.523 0.806 0.888 0.450

–0.727 –0.315 –0.157 –0.558

2.63 3.28

1.67 1.48

0.907 0.380

–0.196 –0.467

3.16

1.78

0.347

–1.067 (continued)



Table 1

(continued)

Item (33) It’s easy for me to be affectionate with my partner. (34) I worry that I won’t measure up to other people. (35) My partner really understands me and my needs. (R) (36) My partner only seems to notice me when I’m angry.

M

SD

Skewness

2.42 3.75 2.92 2.23

1.41 1.86 1.40 1.39

1.106 –0.081 0.570 1.128

Kurtosis 0.568 –1.279 –0.034 0.701

Note: (R) indicates a reverse-scored item.

Table 2 Fit Statistics for the Adult Attachment Models 2

Model

MLχ

One-factor model Two-factor model

2,992.68 1,636.09

df 594 593

SRMR .11 .072

CFI .91 .96

RMSEA a

.16 (.16, .16) .073 (.069, .076)a

∆χ

2

— 1,356.59*

∆df — 1

2

Note: N = 397. MLχ = maximum likelihood chi-square; SRMR = standardized root mean square residual; CFI = comparative fit index; RMSEA = root mean square error of approximation; ∆χ2 = change in chisquare; ∆df = change in degrees of freedom. a. 90% confidence intervals for the RMSEA. *p < .001.

for the two-factor model fell below .08 (SRMR = .072), and the CFI was greater than .95 (CFI = .96). However, the RMSEA did not fall below .06 (RMSEA = .073). This latter finding suggested the possible presence of model misspecification. Still, the overall fit of the two-factor model appeared adequate, and the latent factors were moderately correlated, r = .51.

CFA: Parameter Estimates Given adequate model fit, parameter estimates were examined. All unstandardized factor pattern coefficients for the items were statistically significant at p < .05. Additionally, all standardized factor pattern coefficients were moderate to high (i.e., ranged from .46 to .76). Subsequent squaring of these standardized estimates yielded an R2 value, or variance accounted for in an item by the latent factor (see Table 3). Finally, the structure coefficients, which provide a means to estimate the indirect effects of a factor on a given item whose factor pattern coefficient has been set to zero, were moderate (see Table 3). R2 values ranged from .21 to .58. According to Hair, Anderson, Tatham, and Black (1995), one should hope to find that at least 50% of the variance in a given item is accounted for by the latent factor for which it serves as an indicator. Therefore, investigation of R2 values may aid in identifying items that contain undesirably large amounts 2 of residual variance. Examination of the standardized error variances, or 1 – R , provides a similar assessment. Inspection of these individual residuals revealed that cer-



Table 3 Standardized Factor Pattern and Structure Coefficients, and Variance Accounted for in the Items by the Latent Factors for the Two-Factor Experiences in Close Relationships–Revised (ECR-R) Solution 2

Item

Anxiety

Avoidance

R

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36

.00 (.25) .00 (.27) .00 (.29) .00 (.33) .00 (.34) .00 (.34) .00 (.32) .00 (.36) .00 (.29) .00 (.38) .00 (.34) .00 (.37) .00 (.37) .00 (.34) .00 (.32) .00 (.34) .00 (.36) .00 (.34) .73 (.73) .73 (.73) .75 (.75) .76 (.76) .67 (.67) .67 (.67) .65 (.65) .72 (.72) .63 (.63) .52 (.52) .65 (.65) .52 (.52) .53 (.53) .48 (.48) .52 (.52) .49 (.49) .58 (.58) .46 (.46)

.49 (.49) .53 (.53) .57 (.57) .64 (.64) .67 (.67) .66 (.66) .62 (.62) .70 (.70) .57 (.57) .73 (.73) .66 (.66) .72 (.72) .73 (.73) .66 (.66) .62 (.62) .66 (.66) .70 (.70) .67 (.67) .00 (.37) .00 (.38) .00 (.39) .00 (.39) .00 (.34) .00 (.34) .00 (.33) .00 (.37) .00 (.32) .00 (.27) .00 (.34) .00 (.27) .00 (.27) .00 (.25) .00 (.27) .00 (.25) .00 (.30) .00 (.24)

.24 .28 .33 .41 .44 .44 .39 .49 .32 .53 .44 .52 .53 .43 .38 .43 .49 .45 .53 .53 .57 .58 .44 .44 .42 .52 .39 .27 .42 .27 .28 .23 .27 .24 .33 .21

Note: N = 397. The standardized parameter estimates are presented first, followed by the structure coefficients in parentheses. R2 represents the variance accounted for in an item by the latent factor for which it serves as an indicator.

tain items had large amounts of unexplained variance. Specifically, Items 20, 24, 26, 28, 30, 32, and 36 (all written to represent the anxiety factor) had large error variances, ranging from .71 to .79. Interestingly, these items seem to address issues not specific



to anxiety. For example, Items 24 and 28 seem to address closeness, as Item 24 reads, “I find that my partner(s) don’t want to get as close as I would like”; and Item 28 reads, “My desire to be very close sometimes scares people away.” Furthermore, Items 32 and 36 seem to address anger, as Item 32 reads, “It makes me mad that I don’t get the affection and support I need from my partner”; and Item 36 reads, “My partner only seems to notice me when I’m angry.” By addressing extraneous content such as this, these items are not effectively representing the latent construct for which they were written.

CFA: Examining Residuals Although overall model fit was adequate, which allowed for interpretation of parameter estimates, it is prudent to investigate areas of misfit, albeit slight or minor. To understand model misfit as potentially indicated by the RMSEA, standardized covariance residuals were examined. Standardized residuals provide information regarding the discrepancy between observed item covariances and model implied item covariances, which lends insight into those relationships between items that are not well reproduced by the model. Although this information effectively provides a guide to areas of misfit in the model, it does not inform why there is model misfit. There were large standardized residuals for Items 13 and 27 (7.24, both written to serve as indicators for the avoidance factor), Items 24 and 28 (6.90, both written to serve as indicators for the anxiety factor), and Items 29 and 31 (7.69, both written to serve as indicators for the avoidance factor). To enhance the information gained about model misfit from the examination of these residuals, modification indices (MIs) were also examined. Although researchers (e.g., MacCallum, Roznowski, & Necowitz, 1992) do not recommend post hoc implementation of MIs, these indices provide a useful guide for identifying misfit. The MIs recommended that items with similar wording should have correlated error terms. These were the same items that had large standardized residuals described above, which is not surprising given that the lack of fit identified by the standardized residuals is precisely what the modifications should alleviate if implemented. For example, the MIs suggested correlating error terms between Items 13 and 27 (decrease in χ2 = 52.4), both of which addressed anxiety experienced when a romantic partner wished to become close. Additionally, the MIs also recommended correlating 2 error terms between Items 29 and 31 (decrease in χ = 59.2), both of which addressed the capacity to depend on a romantic partner. Although items written to represent the same construct should be highly related, correlated error terms such as these represent a source of additional variance that the items share above and beyond the intended construct. Because this extraneous variance is not modeled, the accuracy of the reproduced covariances is decreased. Finally, the MIs also detected item cross-loadings. For example, the addition of a path from the avoidance factor to Item 20 (written to serve as an indicator for anxiety) would decrease the overall χ2 by 43.8. This suggests that the item is multidimensional, containing variance explained by both the avoidance and the anxiety factors.



Table 4 Reliability, Means, and Standard Deviations of the Experiences in Close Relationships–Revised (ECR-R) Subscales Avoidance Subscale Cronbach’s coefficient α M SD Variance accounted for by the latent factors

a

.927 (.916, .937) b 50.000 18.175 .395

Anxiety Subscale a

.917 (.904, .928) b 57.076 19.351 .387

Note: N = 397. a. 95% confidence intervals for Cronbach’s coefficient alpha, calculated using the central F distribution (Fan & Thompson, 2001). b. Possible range of scores: 18-126.

Notwithstanding these results, general fit for the model was assessed to be adequate, and the study’s additional research questions were therefore pursued. It should be noted, however, that the discussion of issues like the ones above are particularly useful when major scale endeavors, such as rewriting or adding items are undertaken. Further discussion related to this is held for the discussion after additional validity evidence has been evaluated.

Intercorrelations and Internal Consistency of the ECR-R Subscale Scores Means, standard deviations, reliability estimates, and the variance accounted for associated with the ECR-R subscales are presented in Table 4. Cronbach’s coefficient alpha estimates of internal consistency for the ECR-R scores were similar to original research conducted on the ECR. However, the variance accounted for in the items by the latent factors was less than 50% for both subscales. Seeing that the total variance accounted for by the latent factors is a function of the individual R2 values associated with the items, it is not surprising that these values are not high.

Establishing a Nomological Net: Convergent and Divergent Validity Evidence for the ECR-R Scores Relationship with Touch subscales. There were mixed results regarding the predictions of the relationships between the ECR-R scores and scores from the Brennan, Wu, et al. (1998) Touch Scale (see Table 5). First, as predicted, there was a positive relationship between scores from the ECR-R Avoidance subscale and scores from the Touch Avoidance subscale of the Touch Scale. Second, and also as predicted, there was a negative relationship between scores from the ECR-R Avoidance subscale and scores from the Affectionate Proximity subscale of the Touch Scale. Third, and con-



Table 5 Correlations Between the Experiences in Close Relationships–Revised (ECR-R) Subscales and Related Criterion Scales Criterion Touch subscales Touch Avoidance Affectionate Proximity Desire for Touch Safe-Haven Touch UCLA Loneliness Scale SPS PSWQ

ECR-R Anxiety Subscale

ECR-R Avoidance Subscale

NP NP NP NP .528 –.431 .386

.511 –.512 .330 –.412 .368 –.454 NP

Note: N = 370. NP = no prediction hypothesized; SPS = Social Provisions Scale; PSWQ = Penn State Worry Questionnaire.

trary to prediction, there was a positive relationship between scores from the ECR-R Avoidance subscale and scores from the Desire for Touch subscale of the Touch Scale. Finally, and as predicted, there was a negative relationship between scores from the ECR-R Avoidance subscale and scores from the Safe-Haven Touch subscale of the Touch Scale. Relationship with loneliness. Predictions for the relationship between scores from both the Anxiety and Avoidance subscales of the ECR-R and scores from the UCLA Loneliness Scale were supported. Specifically, there was a positive relationship between scores from the ECR-R Anxiety subscale and scores from the UCLA Loneliness Scale. Further, there was a positive relationship between scores from the ECR-R Avoidance subscale and scores from the UCLA Loneliness Scale. Relationship with social support. Predictions for the relationship between scores from the Anxiety and Avoidance Subscales of the ECR-R and scores from the SPS were supported. That is, there was a negative relationship between scores from the ECR-R Anxiety subscale and scores from the SPS. Likewise, there was a negative relationship between scores from the ECR-R Avoidance subscale and scores from the SPS. Relationship with worry. Predictions for the relationship between scores from the ECR-R Anxiety subscale and scores from the PSWQ were also supported. Specifically, there was a positive relationship between scores from the ECR-R Anxiety subscale and scores from the PSWQ. Overall, investigation into the construct’s nomological net provided good validity evidence for the scores from the ECR-R. Specifically, all hypothesized relationships were supported, with the exception of the relationship between scores from the Avoidance subscale of the ECR-R and scores from the Touch Avoidance subscale of



the Brennan, Wu, et al. (1998) measure. This finding is considered further in the discussion.

Discussion The present investigation had three goals: (a) to conduct a CFA on the ECR-R scores to further investigate the hypothesized factor structure; (b) to estimate the internal consistency of the scores for each subscale and evaluate their adequacy; and (c) given support for the scale’s internal structure, to test hypothesized relationships between the two dimensions of attachment and theoretically related constructs to assess validity evidence for the scale scores. To investigate the first research question, a CFA was conducted to test the distinctiveness of the two hypothesized dimensions of anxiety and avoidance. Analysis revealed general support for a two-factor solution, with a moderate disattenuated correlation between the latent factors. Although two out of three fit indices fell within recommended ranges, the RMSEA was slightly above proposed standards. Standardized residuals and MIs were therefore examined to identify areas of misfit. This examination revealed that several items sharing similar wording or subject matter were driving the minor misfit. One might consider removing items from these redundant pairs to reduce the misfit, as item removal of this kind would not only lead to a more parsimonious scale while still covering the breadth of the construct but would also eliminate current sources of error covariation. However, it should be noted that scale changes of this kind would necessitate reevaluating validity evidence for the scores on the new version of the scale. Beyond the issue of fit, there were concerns over explained variance in the ECR-R items. It should be noted that minimal item variance might be explained even when adequate model fit is achieved; therefore, it is important to examine the amount of variance accounted for in the items by the factors. Notably, Items 20 and 36 (both specified to load on the anxiety factor) had particularly low R2 values: .27 and .21, respectively. The correlation matrix of the data also revealed that these items had low, and indiscriminant, correlations with other items on the scale. This issue of unexplained variance in the items should be addressed in one of two ways. First, if the items are representing an extraneous construct that is not relevant to anxiety, then they should be removed. However, if these items truly represent an important dimension of anxiety, then it is necessary to better represent the dimension by adding more items that measure it directly. Modifications of this kind would result in larger R2 values for the items. To that end, researchers may later choose to revise or remove Item 20 if similar issues consistently arise in future empirical work. Although the functioning of particular items such as these undoubtedly raise concern, it should be reiterated that overall fit for the model was quite good. With reference to the second purpose of the study, analysis indicated good internal consistency estimates of scores from both subscales as measured by Cronbach’s coefficient alpha (i.e., above .90). However, variance accounted for in the ECR-R items by



the latent factors was less than .50 for each subscale, suggesting that a majority of variance in the items was left unexplained by the latent factors. Again, this leads one to question whether the items on the scale are efficiently representing the constructs of anxiety and avoidance. Finally, with reference to the third purpose of the study, external domain investigations provided generally good construct validity evidence for the scores. Specifically, all hypothesized relationships were supported except for the relationship between avoidance and the desire for touch. That is, a positive relationship was found between these two constructs. This suggests, contrary to the original hypothesis, that those individuals who are avoidant (and therefore less likely to experience touch) actually desire more touch. This finding may imply that there is a common, base need for touch that needs to be realized, whether one is avoidant or not. However, this notion needs to be investigated further before stating it with any certainty. Specifically, the present study only represents the second empirical effort to relate scores from the Touch Scale and scores from a version of the ECR. Furthermore, in the original Brennan, Wu, et al. (1998) investigation, there was no directional relationship specified between the touch subscales and the original ECR subscales. Rather, the study simply addressed variance accounted for in the touch subscales. Further investigation of this relationship should be pursued to clarify this point.

Implications, Limitations, and Future Research In sum, although we found limited areas of weakness, the ECR-R seems to perform adequately enough to provide preliminary support for inferences from the scores. Still, revisions of the scale should be pursued to improve item content and to increase variance accounted for in the items by the latent constructs. Also, it should be noted that generalizability of our results might be compromised due to the largely homogeneous college sample employed. Therefore, future investigations should first replicate the current study on a more diverse population. Additionally, continuing investigations into the nomological net of adult romantic attachment may be useful. These studies may either broaden the scope of the related constructs examined or may choose to cross-validate those currently used with alternative measures of the same constructs. Finally, as there were substantially more predictions made with reference to the Avoidance subscale of the ECR-R, future work should incorporate more a priori hypotheses about the relationship between scores from the ECR-R Anxiety subscale and theoretically related constructs. Further pursuit of validity investigations of this kind may provide the necessary information to better understand, and ultimately improve, the functioning of this scale.

References Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Patterns of attachment: A psychological study of the strange situation. Hillsdale, NJ: Lawrence Erlbaum. Ainsworth, M. S. (1989). Attachments beyond infancy. American Psychologist, 44, 709-716.


134 Educational and Psychological Measurement American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: Author. Bartholomew, K., & Horowitz, L. M. (1991). Attachment styles among young adults: A test of a fourcategory model. Journal of Personality and Social Psychology, 61, 226-244. Becker, T. E., Billings, R. S., Eveleth, D. M., & Gilbert, N. W. (1997). Validity of scores on three attachment style scales: Exploratory and confirmatory evidence. Educational and Psychological Measurement, 57, 477-493. Benson, J. (1998). Developing a strong program of construct validation: A test anxiety example. Educational Measurement: Issues and Practice, 17, 10-17. Berube, M. S., et al. (Eds.). (1982). The American heritage dictionary (2nd College ed.). Boston: Houghton Mifflin. Bowlby, J. (1973). Attachment and loss: Vol. 2. Separation: Anxiety and anger. New York: Basic Books. Bowlby, J. (1977). The making and breaking of affectional bonds. British Journal of Psychiatry, 130, 201210. Bowlby, J. (1979). The making and breaking of affectional bonds. London: Tavistock. Bowlby, J. (1980). Attachment and loss: Vol. 3. Loss: Sadness and depression. New York: Basic Books. Bowlby, J. (1982). Attachment and loss: Vol. 1. Attachment (2nd ed.). New York: Basic Books. Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-report measurement of adult attachment: An integrative overview. In J. Cassidy & P.R. Shaver (Eds.), Handbook of attachment: Theory, research, and clinical applications (pp. 434-465). New York: Guilford. Brennan, K. A., & Shaver, P. R. (1995). Dimensions of adult attachment, affect regulation and romantic relationship functioning. Personality and Social Psychology Bulletin, 21, 267-283. Brennan, K. A., Wu, S., & Loev, J. (1998). Adult romantic attachment and individual differences in attitudes towards physical contact in the context of adult romantic relationships. In J. Cassidy & P. R. Shaver (Eds.), Handbook of attachment: Theory, research, and clinical applications (pp. 394-428). New York: Guilford. Collins, N. L. (1996). Working models of attachment: Implications for explanation, emotion, and behavior. Journal of Personality and Social Psychology, 71, 810-832. Collins, N. L., & Read, S. J. (1990). Adult attachment, working models, and relationship quality in dating couples. Journal of Personality and Social Psychology, 58, 644-663. Cutrona, C. E., & Russell, D. W. (1987). The provisions of social relationships and adaptation to stress. Advances in Personal Relationships, 1, 37-67. Davis, M. H., Morris, M. M., & Kraus, L. A. (1998). Relationship-specific and global perceptions of social support: Associations with well-being and attachment. Journal of Personality and Social Psychology, 74, 468-481. Fairchild, A. J., Pastor, D. A., & Brennan, K. A. (2003, April). Exploratory factor analysis of an adult attachment measure. Poster presented at the Virginia Psychological Association Spring Convention, Tysons Corner, VA. Fan, X., & Thompson, B. (2001). Confidence intervals around score reliability coefficients, please: An EPM guidelines editorial. Educational and Psychological Measurement, 61, 517-531. Feeney, J., Noller, P., & Hanrahan, M. (1994). Assessing adult attachment. In M. B. Sperling & W. H. Berman (Eds.), Attachment in adults: Clinical and developmental perspectives (pp. 128-152). New York: Guilford. Florian, V., Mikulincer, M., & Bucholtz, L. (1995). Effects of adult attachment style on the perception and search for social support. Journal of Psychology, 129, 665-676. Fraley, R. C., & Waller, N. G. (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. C., Waller, N. G., & Brennan, K. A. (2000). An item response theory analysis of self-reports measures of adult attachment. Journal of Personality and Social Psychology, 78, 350-365. Gana, K., Bettina, M., & Canouet, M. (2001). Worry and anxiety: Is there a causal relationship? Psychopathology, 34, 221-229.


Fairchild, Finney / Validity Evidence for the ECR-R Questionnaire 135 Griffin, D. W., & Bartholomew, K. (1994). The metaphysics of measurement: The case of adult attachment. Advances in Personal Relationships, 5, 17-52. Hair, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (1995). Multivariate data analysis with readings (4th ed.). Upper Saddle River, NJ: Prentice Hall. Hawkins, M. J. (1995). Anxiety in relation to social support in a college population. Journal of College Student Psychotherapy, 9, 79-88. Hazan, C., & Shaver, P. R. (1987). Romantic love conceptualized as an attachment process. Journal of Personality and Social Psychology, 52, 511-524. Hazan, C., & Shaver, P. R. (1994). Attachment as an organizational framework for research on close relationships. Psychological Inquiry, 5, 1-22. Hu, L., & Bentler, P. M. (1995). Evaluating model fit. In R. H. Hoyle (Ed.), Structural equation modeling: Concepts, issues and applications (pp. 76-99). Thousand Oaks, CA: Sage. Hu, L., & Bentler, P. M. (1998). Fit indices in covariance structure modeling: Sensitivity to underparameterized model misspecification. Psychological Methods, 3, 424-453. Hu, L., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling, 6, 1-55. Jöreskog, K. G., & Sörbom, D. (1993). LISREL 8: Users’ reference guide. Chicago: Scientific Software International. Kline, R. B. (1998). Principles and practice of structural equation modeling. New York: Guilford. MacCallum, R., Roznowski, M., & Necowitz, L. (1992). Model modifications in covariance structure analysis: The problem of capitalization on chance. Psychological Bulletin, 111, 490-504. Man, K., & Hamid, P. N. (1998). The relationship between attachment prototypes, self-esteem, loneliness and causal attributions in Chinese trainee teachers. Personality and Individual Differences, 24, 357-371. Meyer, T. J., Miller, M. L., Metzger, R. L., & Borkovec, T. D. (1990). Development and validation of the Penn State worry questionnaire. Behavior Research and Therapy, 28, 487-495. Morrison, T. L., Goodlin-Jones, B. L., & Urquiza, A. J. (1997). Attachment and the representation of intimate relationships in adulthood. Journal of Psychology, 131, 57-71. Russell, D. W. (1996). UCLA Loneliness Scale (Version 3): Reliability, validity and factor structure. Journal of Personality Assessment, 66, 20-40. Simpson, J. A. (1990). Influence of attachment styles on romantic relationships. Journal of Personality and Social Psychology, 59, 971-980. West, S. G., Finch, J. F., & Curran, P. J. (1995). Structural equation models with nonnormal variables: Problems and remedies. In R. H. Hoyle (Ed.), Structural equation modeling: Concepts, issues and applications (pp. 76-99). Thousand Oaks, CA: Sage.
