Introduction

In this study we examined the factor structure of the “Experiences in Close Relationships” instrument (Brennan et al. 1998; ECR) for assessing adult attachment among a clinical sample by conducting a series of exploratory factor analyses (EFAs). The factor structure of the ECR was originally determined using results from a sample of 1,086 undergraduates with a median age of 18 to determine the particular items that best measure the construct (Brennan et al. 1998). The measure demonstrated high reliability and strong validity among the targeted population of college undergraduates, however, the factor structure had yet to be examined among participants involved in therapy. It is unclear if any results obtained using the ECR extrapolate to populations other than the original sample. Our findings support the factor structure of the ECR for assessment of attachment anxiety and avoidance among a more diverse, clinical population.

The ECR has been used to examine the influence of attachment on various clinical variables, such as emotional reactivity and depression (Lopez et al. 2001; Wei et al. 2005). However, threats to internal and construct validity are posed when using the ECR to assess attachment among a clinical sample, as the factor structure for clinical participants has not been verified previously. Specifically, items that do not properly load onto designated factors of the measure (i.e., anxiety, avoidance) may suggest the presence of an underlying construct or inconsistency in item interpretation. Either circumstance could contribute unexplained variance, potentially creating a bias that influences the results. This study identifies which items assess adult attachment subscales of anxiety and avoidance using a clinical sample. By exploring which items accurately identify components of attachment anxiety and avoidance among a clinical population, outcome variables may be more accurately assessed in future studies. Additionally, the present study includes a separate analysis of the attachment subscales for men and women to identify gender differences in responding to items in the ECR.

Inquiry into the relational or interactive nature of attachment (Whiffen 2005) underscores the need to examine the factor structure of the ECR, as studies indicate the influence of attachment in relationships varies by gender. For example, when comparing the influence of attachment on symptom distress, anxious women married to avoidant men did not experience increased attachment anxiety or depressive symptoms. In contrast, anxious husbands were found to be more vulnerable to depression in the event that their spouses demonstrated higher levels of avoidance. Although attachment has been studied extensively as an influencing factor of relational variables, such as satisfaction, adjustment, and distress (Frei and Shaver 2002; Scott and Cordova 2002; Treboux et al. 2004; Wampler et al. 2003), original results were not compared between men and women (Brennan et al. 1998). The differences found between men’s and women’s attachment influences on mental health symptoms further highlight the need to examine the assessment of attachment separately by gender among a clinical sample.

Methods

Participants

The sample consisted of 1,138 individuals participating in therapy from 2003 to 2008. The majority of men and women participating in the study were Caucasian (62.4%, n = 355—men; 69.4%, n = 395—women); 10.5% of men (n = 60) and 11.2% of women (n = 64) in the sample identified as African-American; 1.9% of men (n = 11) and women (n = 11) identified as Latino; and 2.8% of men (n = 16) and 3.2% of women (n = 18) identified as “other”. In terms of education, 30.4% of the men (n = 173) and 26.5% of the women (n = 151) in the sample reported a high school diploma or GED as their highest level of education; 14.5% of the men (n = 83) and 15.5% of the women (n = 88) received either an associate’s degree or vocational-technical school training; 25.5% of the men (n = 145) and 29.4% of the women (n = 167) received a bachelor’s degree; 11.4% of the men (n = 65) and 13.5% of the women (n = 77) received a master’s degree. Relative to income, 35.4% of the men (n = 201) and 38.7% of the women (n = 220) reported earning less than $20,000 as their yearly income, while 27.2% of the men (n = 155) and 31.6% of the women (n = 180) reported $20,001 to $40,000 for their yearly income. Finally, 22% of the men (n = 125) and 20.7% of the women (n = 118) in the sample received more than $40,000 for their yearly income.

Participants received therapy from one of two marriage and family therapy training clinics that serve members of the community. The two training clinics from which data were collected were located in the southeastern United States. Both clinics are associated with COAMFTE accredited marriage and family therapy programs and have been approved by the university Human Subjects office for research participation. However, one clinic was a master’s level program, whereas the other was a doctoral level program. At each of the participating clinics, a series of clinical assessments are distributed to participants for a variety of clinical concerns. At the first therapy session, clients are invited to sign a consent form for research participation that allows the completed clinical assessments to be analyzed for research purposes. Clinicians explain to all clients that they will be assigned a random client number and no identifying information will be included for research purposes.

Experiences in Close Relationships Measure

The measure used to examine adult attachment style was the Experiences in Close Relationships questionnaire (Brennan et al. 1998). The measure is a 36-item, self-report scale (See Appendix), which consists of two 18-item subscales, Anxiety and Avoidance. Responses are on a 7-point Likert scale, ranging from “Disagree Strongly” to “Agree Strongly”. The original authors constructed the two 18-item subscales following an exploratory factor analysis of 323 items derived from virtually every other self-report adult romantic attachment measure, supporting the construct validity of the measure. Brennan and colleagues explain that the four clusters found in the original EFA revealed four distinct groups, representing patterns similar to the model of attachment described by Bartholomew and Horowitz (1991). The two scales were found to be nearly uncorrelated (r = .11), suggesting that the measure captures two separate, underlying dimensions of adult attachment. Original alpha scores for both the avoidance (α = .94) and anxiety (α = .91) subscales indicate high reliability of the measure (Brennan et al.).

The ECR produced stronger, more specific measurement precision compared to the Relationship Questionnaire (Bartholomew and Horowitz 1991) and Collins’ and Read’s (1990) Adult Attachment Scale (Fraley et al. 2000). Further, the ECR provides two continuous subscale scores on each attachment dimension, highlighting the difference in anxiety and avoidance scores among individuals in each attachment category (i.e., secure, fearful, preoccupied, dismissing). Subscale scores more accurately discriminated participants through different degrees of attachment insecurity, specifically the distinction of fearful and dismissing styles (Brennan et al. 1998). The authors also suggest that the ECR is more conservative than Bartholomew’s and Horowitz’s (1991) measure in classifying a person as secure, which has led to statistically stronger results. As such, assessing adult attachment style with the ECR allows for more precise discrimination than do any other prior measures.

Analyses

An exploratory factor analysis (EFA) is conducted to determine the consistency of factors within an instrument to detect any underlying dimensions, as factor structures may differ across samples (Floyd and Widaman 1995). Preliminary analyses included mean and standard deviation (SD) of the sample scores, skewness, kurtosis, and intercorrelations for each item of the anxiety and avoidance subscales. Primary analyses consisted of four separate EFAs; anxiety and avoidance subscales were analyzed separately for male and female participants using a Varimax rotation method. Specifically, a principal component analysis was conducted, which is one type of EFA used to determine underlying components of the measure examined. Floyd and Widaman (1995) support the use of a principal component analysis for the purpose of eliminating extraneous sources of variance in a scale by removing items that do not load onto the first component or factor.

The first factor extracted is that which accounts for the maximum variance in the observed variables. Applied to this study, a principal component analysis was conducted to account for the most variance in attachment dimensions with the fewest items. There are no strict criteria for selecting a particular number of factors to examine when conducting an EFA. However, Gorsuch (1993) recommends evaluating the scree plot to determine the elbow, where the eigenvalues decrease dramatically. As such, factors for each attachment subscale (i.e., anxiety, avoidance) were retained based on substantial decline in total variance explained by each factor in the scree plot. Among the two attachment subscales, each item was examined separately based on two criteria. Items that (1) demonstrate a factor loading of <.4, or (2) load on more than one factor of the anxiety or avoidance subscales were not retained for the final inventories.

Results

Skewness values are considered acceptable below an absolute value of two, while kurtosis values are considered acceptable if they do not exceed an absolute value of seven (Gorsuch 1983). The distribution of the items on the ECR indicates that the majority of the items for both the anxiety and the avoidance subscales were within acceptable parameters for skewness and kurtosis, with one exception. Skewness and kurtosis for item 10 among male participants exceeded the acceptable values. Therefore, item 10 was not included in the primary analyses for male participants.

Items in the correlation matrix that were conceptually related exhibited higher levels of correlation (e.g., anxiety subscale items). On the other hand, items that were not significantly correlated to the remaining items exhibited lower levels of correlation, supporting the internal consistency of the measure. Repeated non-significance among particular items in the measure suggests a lack of consistency in the responses given by participants. One explanation for such a trend is the presence of an underlying component that is present for clinical participants, underscoring the need for an EFA of the measure.

Reliability of the anxiety and avoidance subscales of the ECR for both male and female responses was established using Cronbach’s α to estimate internal consistency. Results indicated that reliability among men’s responses on the anxiety subscale was α = .91, and α = .90 on the avoidance subscale. Women’s responses indicate a reliability score of α = .90 on both the anxiety and avoidance subscales.

Anxiety Subscale

Primary analyses for the anxiety subscale of male participants yielded the results indicated in Table 1. When the number of factors was limited to three, 56.7% of the variance was explained by the factors as indicated in the scree plot. Consistent with the original EFA, one factor was retained. Additionally, no items were eliminated due to insufficient factor loading (i.e., <.40), which supports the Brennan et al. (1998) findings of strong reliability and validity of the original measure. Items 10, 12, 16, and 26 loaded on multiple factors, suggesting that the items also may measure an additional construct captured by the anxiety subscale among male clinical participants.

Table 1 ECR factor loadings—men’s anxiety subscale
Full size table

The EFA of the anxiety subscale was examined separately for females. Analyses for the female participants yielded the results indicated in Table 2. When the number of factors was limited to three, 53.6% of the variance was explained by the factors as indicated by the scree plot. Consistent with the men, no items were eliminated due to insufficient factor loading (i.e., <.40) among women. Items 2, 14, 30, 32, and 36 loaded on multiple factors, suggesting that the items also may measure an additional construct captured by the items among female clinical participants. Item 22 loaded sufficiently onto Factor 3, possibly identifying a separate construct within the subscale. However, there were no other items that loaded sufficiently onto the factor, so the item was removed in subsequent analyses.

Table 2 ECR factor loadings—women’s anxiety subscale
Full size table

Avoidance Subscale

The EFA of the avoidance subscale also was examined separately for males and females. Results for the male participants are presented in Table 3. When the number of factors was limited to three, 53.8% of the variance was explained by the factors indicated in the scree plot. Item 19 was eliminated due to insufficient factor loading (i.e., <.40), indicating that there was little consistency of responses to this item among the men in the sample. Items 21, 29, 31, 33 and 35 loaded on multiple factors, suggesting the items also may measure an additional construct captured by the avoidance subscale among male clinical participants.

Table 3 ECR factor loadings—men’s avoidance subscale
Full size table

Results for the avoidance subscale among female participants are presented in Table 4. The number of factors was limited to three, based on 56.7% explained variance of the factors indicated in the scree plot. Item 33 was eliminated due to insufficient factor loading (i.e., <.40), indicating that there was little consistency of responses to the item among women in the sample. Items 21, 25, 27 29, 31, and 35 loaded on multiple factors, suggesting the items also may measure an additional construct captured by the items among female clinical participants.

Table 4 ECR factor loadings—women’s avoidance subscale
Full size table

Reliability of the anxiety and avoidance subscales following the EFA were repeated for both the male and female responses using Cronbach’s α to estimate internal consistency of the revised subscales. Results indicated that reliability remained high after removing the items that did not load properly based on the established criteria. Specifically, men’s responses on the revised anxiety and avoidance subscales were α = .90. Women’s responses indicate a reliability score of α = .85 on the anxiety subscale and α = .90 on the avoidance subscale.

Discussion

One explanation for multiple factor loadings of similarly worded items is inconsistent interpretation of the items among respondents across different samples (Floyd and Widaman 1995). Applied to the present study, a clinical population may interpret items differently than a non-clinical sample, which can affect both research and clinical findings. The ECR (Brennan et al. 1998) measure of adult attachment has demonstrated strong validity and reliability among the original sample of college students, however, a factor analysis of the measure had not been conducted previously with a clinical population. Different factor structures found among the clinical sample suggest that the items used to measure attachment anxiety or avoidance may have meanings for participants that differ from those of the original sample. In contrast to the original sample, participants in this sample included men and women over the original sample median age of 18, a greater variety of education levels (i.e., college undergraduates), and with various degrees of mental health symptom distress. Additionally, the analysis revealed themes based on wording of the items that differed for men and women.

There were two noteworthy differences between anxiety for men and women. Items related to worry (i.e., 2, 4, 6, 8, 14, 22), partner disclosure (i.e., 18, 20, 24), and time spent together (i.e., 30, 32, 36) all loaded onto the anxiety factor for men. The anxiety subscale for women revealed similar themes, including worry (i.e., 4, 6, 8), partner disclosure (i.e., 18, 20, 24), and personal desire for closeness (i.e., 10, 12, 16, 26). Personal desire for closeness items were associated with attachment anxiety for women in therapy, but not for men. The time spent together theme loaded onto anxiety for men, but not women, indicating that amount of time spent with a partner was a contributing theme to attachment anxiety among men in therapy.

In contrast to the anxiety subscale, there was a close similarity of themes for both male and female partners on the avoidance subscale. Two themes were present for avoidance among male participants, disclosure (i.e., 1, 9, 15, 25, 27) and comfort with closeness (i.e., 3, 5, 7, 11, 13, 17, 23). Avoidance among women displayed similar themes, substituting only a few items for disclosure (i.e., 1, 9, 15) and comfort with closeness (i.e., 3, 5, 7, 11, 13, 17, 19, 23). However, future studies that include the ECR (Brennan et al. 1998) may benefit from removing dependence (i.e., 21, 29, 31, 35) items, as they may contribute unexplained variance to the model being tested. The themes revealed among the anxiety and avoidance subscales have implications for future studies using the ECR.

Research Implications

The goal of this study was to examine the underlying factor structure of the ECR for use with a clinical sample. Differences were revealed in the factor structures between the original sample used by Brennan et al. (1998) and a clinical sample by identifying items in each subscale that did not load properly. Removing these particular items allows the researcher to draw more clear inferences from his or her study, thereby strengthening the conclusion validity of clinical studies using the ECR. For example, one study found a significant association between high anxiety and avoidance scores on the ECR and various mental health diagnoses among a clinical sample (Riggs et al. 2007). However, removing the aforementioned items from the subscales would allow the researchers to make stronger inferences about the relationship between attachment of clinical participants and mental health diagnoses. Further, Riggs and colleagues found no gender differences in the association between anxiety or avoidance and mental health symptoms among the sample. Findings from the present study could increase the validity of results and expand future studies by removing unexplained variance of ECR scores of male and female clinical participants.

By removing sources of extraneous variance from the subscales, construct validity is also strengthened when using the ECR. Messick (1995) notes that “construct-irrelevant variance” (p. 742) in psychological assessment contributes excess reliable variance associated with other, distinct constructs in a manner that is irrelevant to the interpreted construct. Applied to the present study, items of the ECR subscales that loaded on multiple factors may indicate the presence of an underlying component or construct that is present in the interpretation of the items among a clinical sample. One method of identifying such constructs is to explore the wording used in the multiple-loading items to reveal themes that may be relevant to attachment assessment among clinical samples. Although this study made contributions to couple therapy research, clinical implications also may be drawn for use with couples in therapy.

Clinical Implications

Results from the present study offer implications for developing individual and couple attachment-informed interventions with clients participating in therapy. Specifically, individuals experiencing high attachment anxiety often exhibit particular behaviors that contribute to distress, such as heightened emotion and negative thought processes (Bifulco et al. 2002; Hankin et al. 2005; Wampler et al. 2003). Clinical interventions may be implemented with such clients to highlight, block, or alter negative thought processes and heightened emotion surrounding the corresponding attachment themes (i.e., time spent together, desire for personal closeness). For example, a man who feels his female partner does not spend enough time with him throughout the week may ruminate on negative thoughts about being unloved, which contribute to distress. Efforts to block the rumination on thoughts consistent with the theme (i.e., time spent together) could aid in alleviating distress. Couple interventions that are informed by the contrasting themes revealed between men and women also may be constructed.

Although there was consistency among the avoidance themes, there were noteworthy differences between men and women for anxiety. The contrasting anxiety themes described above may be used to inform clinical practice by constructing appropriate interventions to decrease attachment anxiety. Men who were high on attachment anxiety experienced distress if the need for “time spent together” was not met. In contrast, women high on anxiety experienced distress when “desire for personal closeness” was not met. Based on these findings, assignments or interventions developed in therapy may be more specific to both partners by attending to the separate themes. “Date nights” are a common homework assignment for couples between sessions. Such an intervention may address the need of “time spent together” for a man experiencing anxiety, however, the woman’s “desire for closeness” may not be incorporated into the assignment. The contrasting attachment themes revealed between men and women in this study underscore the significance of couple interventions that attend to both partners’ attachment concerns in order for them to be effective for both partners. At the same time, despite the contributions of the present study, it is important to interpret these results with caution as this study had several limitations.

Limitations

Although this study aims to inform research on couple therapy, it contains several threats to external validity. Results from this study may be generalized only to a particular portion of the population due to the limited variability of the sample. Participants included in the sample were predominantly Caucasian. Therefore, results may not apply to samples that predominantly consist of different racial/ethnic participants. Further, the overall sample was highly educated, obtaining some amount of education beyond a high school diploma. The majority of the sample also reported an annual income above $20,000. Future studies should include couples with greater demographic variability (e.g., racial minority participants, low SES couples) to strengthen the external validity of the results. As a result, the generalization of results from this study is limited to participants similar to those described in the sample.

The present study also contains a threat to construct validity, as the ability of the ECR to accurately assess various attachment styles has come into question. The ECR has shown inaccurate assessment of those with low anxiety and avoidance (i.e., secure style), despite superior measurement of attachment in comparison to other self-report measures (Fraley et al. 2000). Fraley and colleagues suggest that averaging the responses of the ECR subscales limits the ability to identify changes in secure styles over time. Said differently, the measure more accurately assesses insecure styles of attachment (i.e., high anxiety, high avoidance), as opposed to secure attachment over time. It follows that participants scoring low on anxiety and avoidance subscales (i.e., secure style) may not have been accurately identified from insecure styles due to the averaging of subscale item responses. Despite these limitations, the results obtained may be used to inform further attachment research and clinical interventions.

Conclusion

Findings from this study have made several contributions to the clinical and adult attachment literature. In particular, this study has provided the underlying factor structure of attachment anxiety and avoidance measurement separately for men and women in therapy. The attachment literature also was expanded by removing possible contributions of unexplained variance in studies that include adult attachment of couples in therapy. Future studies should attempt to include participants with greater diversity, specifically in the areas of race/ethnicity, education level, SES, and region of the country, so that results may be applicable to a broader clinical population. Further, this sample included solely heterosexual clients, as there were not enough GLBT couples to include in the analyses. Therefore, a direction for future research would be to replicate the study among GLBT clients. Despite the aforementioned areas of concern, this study offers support for the assessment of adult attachment by both researchers and clinicians working with men and women in therapy.