The amount of research into adult attachment has grown exponentially over the past 25 years, and the issue of adult attachment has become increasingly important in medical research, since attachment impacts individuals on a variety of levels, such as personality, coping styles, relationships and health, plus is a predisposing factor for a variety of symptoms such as anxiety, depression and personality disorders, and acts as a working alliance in psychotherapy [1-9]. A number of adult attachment measures have been developed, whether clinician-administered or self-reporting, and a recent review of adult attachment measures showed that at least seven interview methods and 22 self-report questionnaires have been used [10].

The Experiences of Close Relationships (ECR) measure is one of the self-report measures used widely in clinical settings, with clinical samples. The ECR, a 36-item questionnaire based on two dimensions: anxiety and avoidance, was developed by Brennan, Clark et al. [11], and later the original 323-item dataset was re-analyzed by Fraley [12] using item response theory. The self-report ECR is said to be meaningfully related to personality and psychopathology, and can be usefully applied on psychiatric populations [12]. The revised version (ECR-R) contains 18 items which assess romantic attachment anxiety and 18 items which assess attachment avoidance. Item answers form a 7-point Likert-type rating scale, ranging from 1 (disagree strongly) to 7 (agree strongly).

Scores on the anxiety and avoidance scales can still be used to classify people into the four adult attachment styles as developed by Bartholomew and Horowitz [13-15], based on thoughts about oneself (anxiety dimension) and about others (avoidant dimension). The anxiety and avoidance combinations create four attachment styles, these being: secure (low anxiety, low avoidance), preoccupied (high anxiety, low avoidance), fearful (high anxiety, high avoidance) and dismissing (low anxiety, high avoidance).

The ECR-R has been tested for its psychometric properties and found to have an adequate model fit with an Anglo-Saxon sample, plus to be acceptable for non Anglo-Saxon populations [16-19]. Wongpakaran et al. [20] have used a Thai version to test for reliability and validity, and produced the same result; however, it has been suggested that some items contributing to the slight areas of misfit in the model should be deleted in the 36-item version.

Although the ECR appears to be a highly reliable and a valid instrument to use when assessing adult attachment, it has mostly been used with non-clinical samples. When it comes to clinical samples, especially psychiatric patients, such a long questionnaire might easily have an adverse effect upon reliability; therefore, a shorter version needs to be developed. Recently, Wei et al. [21] developed a briefer version of ECR based on the original; the shorter version is comprised of twelve items and has been found to be a valid and reliable when compared to the original ECR; however, a later version of ECR-R - with some items different from the original - has never been developed in a shorter form. In addition, based on our previous study, it has been suggested that some items should be revised or deleted entirely [20]; therefore, the purpose of this study is to seek to revise the Thai version of the ECR-R into a shorter form, while attempting to maintain its construct validity. Exploratory factor analysis was used to determine appropriate items, and later confirmatory analysis methods were used to ensure the validity of the new version of the scale. Criterion validity was examined by finding the correlation between the ECR-R-18 and other self-report measures, such as Rosenberg’s self-esteem scale (RSES) [22], the Perceived Stress Scale [23], and the UCLA Loneliness Scale [24]. We hypothesized that the anxiety sub-scale would be positively correlated with the Perceived Stress Scale, and negatively correlated with the Self-Esteem Scale, whereas attachment avoidance was expected to have a positive correlation with the Loneliness Scale. As far as we are concerned, this study is the first developed to investigate the effectiveness of the shorter version of the ECR-R, as well as the first to compare its effectiveness with both clinical and non-clinical samples.

This study project was approved by the Ethics Committee at the Faculty of Medicine, Chiang Mai University. According to Comrey, five to ten times the number of items is deemed to be a suitable sample size to carry out confirmatory factor analysis [25]; therefore, a convenience sample of 531 for non-clinical and 171 for clinical samples was determined to be appropriate for this study.

A non-clinical sample of 531 people was recruited from the relatives of those accompanying patients, plus from a group of undergraduate students, and hospital personnel. They were invited to participate in the study, and offered a participant information sheet (PIS) and informed consent form by a research assistant, plus were provided with a pack containing the PIS, questionnaires and an informed consent form. The process was carried out on a voluntary basis. No compensation was given to members of the group for their participation, among whom 58% were female aged 18 to 58 years old (M= 39.56 years, SD=8.05 years).

In the clinical sample, we recruited 171 patients from Maharaj Nakorn Chiang Mai Hospital Psychiatry Outpatients Service. The clinically stable patients were invited to participate in the study, were offered a PIS and informed consent form by a research assistant. The informed consent form covered the initial testing and four-week retesting activities. The inclusion criteria were 1) at least 18 years old, and 2) to be able to complete questionnaires. Potential participants were excluded if they revealed: 1) active psychosis 2) present episode of bipolar, manic episodes or severe depression, 3) organic mental disorders, 4) active conditions that required medical attention regardless of the cause; for example, active suicidal behavior, being delirious, intoxicated or withdrawn, and 5) any condition that required hospitalization. Among the recruited patients, 60% were female, aged 18 to 74 years old (M= 42.11 years, SD=16.08 years). According to the DSM-IV-TR, 39% suffered from depressive disorders, 10% had problems with substance abuse/dependence, and the rest suffered from anxiety disorders, somatoform disorders and psychosomatic disorders. Regarding re-testing, it was found that 86% of all participants performed the second round of questions for accuracy.

In terms of descriptive data, there was a clear distinction between the data provided by the two samples. As expected, the patient group appeared to have a higher level of insecure attachment (e.g., preoccupied, fearful and dismissing) than the non-patient group. A further important point was whether the measurement construct was the same between the two groups, and the results shows that it was not. Item 1 (‘I prefer not to show a partner how I feel deep down’) was found to be problematic and seemed to be responsible for the discrepancy since it loads on the other factors in the patient sample. This item was also found to be double-loading in Wei’s study and was not thus chosen for the short ECR [21].

Overall, Cronbach’s alpha was acceptable for both groups but was lower in the clinical group (0.84 for the non-clinical sample; 0.77 for the clinical sample), and was lower in the avoidance sub-scale than in the anxiety scale, especially in the patient group. Notably, most items on the avoidance sub-scale were negatively worded, indicating that the responses were more sensitive to negatively worded questions, especially items containing ‘not’ in the sentence. This is supported by Colosi [34] who found that negative phrases or sentences in questions bring about inconsistency in terms of responses, including a higher frequency of ’don’t know’ responses, and that this impacts more upon clinical than non-clinical groups. The fact that the ECR-R-18 did not achieve an excellent model fit is because there was a high proportion of negatively worded items when compared to positively worded items (11 to 7), giving a greater chance of a low reliability. In general circumstances, negatively worded items can give lower reliability [35-37] but the situation could be worse for a clinical sample, as there may be interference from transient or pseudo-cognitive impairments related to high anxiety or poor concentration, and this may limit rating at that moment when compared to a non-clinical sample. The fact that in this study Item 1 was found to have the lowest quality might be because it is a question for which respondents are not yet aware of its true meaning. All in all, the concept of creating negatively worded items for fear of an acquiescent response bias might be cautiously applied to this sample. Modifications to this item, such as changing it from negative to positive wording and having the revised version tested on a separate sample, should be carried out before using it with clinical patients, plus if ECR-18 is to be used with a clinical sample, we also suggest dropping Item 1 from the calculation of the avoidance score (take the sum of 8 items and divide it by 8 to give the mean of the avoidance score).

In terms of factor structure, this shorter version of ECR-R was generally better than the original one in terms of model fit; ECR-R-18 yielded acceptable fit indices as all of the scores were more than 0.90, except for AGFI. This may not meet the perfect criteria of Hu and Bentler, but was generally an acceptable fit. In terms of the bad fit indices, ECR-R-18 demonstrated a good model fit here also, and overall it was comparable or slightly better than the original 36-question ECR-R (Thai version) (CFI = 0.95, NFI = 0.81, RMSEA = 0.04 and SRMR = 0.10) [20].

Measurement invariance tests showed that both constructs were significantly different, indicating that this measurement, when tested on a non-clinical sample cannot be simply used for a clinical sample and employ similar interpretations. These results suggest that alternative scoring methods may need to be considered when using ECR-R-18 data, otherwise some items should be modified, since the measurements that contain negatively worded items not only harm the reliability but also the predictive validity of the measurements [38].

As expected, the anxiety scale showed a significant, negative correlation with the perceived stress scale and the self-esteem scale, whilst avoidance was found to have a positive correlation with the loneliness score highlighting its concurrent validity. The test-retest reliability of ECR-R-18 was also acceptable, indicating its stability over time in this type of group.