Bowlby’s attachment theory (1980/1993) attempts to explain the tendency for human beings to form certain selective, stable, and enduring emotional ties; when such ties are broken, intense emotional suffering ensues. Though this theory was originally developed to explain primary attachment between children and their caregivers (Bowlby, 1969/1998), in the ’80s, research was extended into adulthood, and adult attachment came to mean the intimate bond one forms with his or her romantic partner (Hazan & Shaver, Reference Hazan and Shaver1987). Since then, adult attachment in romantic relationships has been the subject of numerous studies (for a review, see Mikulincer & Shaver, Reference Mikulincer and Shaver2007).
Research on adult romantic attachment, aims to study beliefs about people’s behavior, feelings, and ways of behaving in romantic relationships (Shaver, Belsky, & Brennan, Reference Shaver, Belsky and Brennan2000). Two dimensions of adult romantic attachment have been identified by means of self-report questionnaires: avoidance and anxiety (Brennan et al., Reference Brennan, Clark, Shaver, Simpson and Rholes1998). The avoidance dimension captures the extent to which one rejects or avoids intimacy and physical and psychological closeness with the other, and how comfortable he or she feels depending on and being supported by others. The anxiety dimension is the extent to which one worries about being abandoned or rejected, has low self-esteem, and undervalues him or herself (Obegi, Morrison, & Shaver, Reference Obegi, Morrison and Shaver2004).The two scales combine to determine a person’s security or insecurity of attachment such that individuals with low anxiety and avoidance present with secure adult attachment (Bartholomew & Horowitz, Reference Bartholomew and Horowitz1991). Secure romantic attachment is characterized by having positive models of oneself, others, and romantic relationships (Olsson, Sorebo, & Dahl, Reference Olsson, Sorebo and Dahl2010).
As Alonso-Arbiol and her colleagues (2010), argue in one study, the Experience in Close Relationships questionnaire -ECR- (Brennan et al., Reference Brennan, Clark, Shaver, Simpson and Rholes1998), which evaluates the two dimensions of adult attachment (Shaver & Mikulincer, Reference Shaver and Mikulincer2002), is among the tests most widely used in studies of adult romantic attachment conducted in the last 10 years. Likewise, an analysis conducted by Fraley, Waller, and Brennan (Reference Fraley, Waller and Brennan2000) conveys that of the self-report questionnaires most often employed to assess adult attachment, the ECR has the fewest limitations and best psychometric properties, justifying its vast use in an array of studies in the fields of social, personality, and clinical psychology. These studies have allowed researchers to examine connections between attachment and several variables, including self-esteem (e.g., Cassidy, Ziv, Mehta, & Feeney, Reference Cassidy, Ziv, Mehta and Feeney2003), coping skills (e.g., Lopez & Gormley, Reference Lopez and Gormley2002), and the ability to maintain intimacy and commit oneself to a romantic relationship (Treboux, Crowell, & Waters, Reference Treboux, Crowell and Waters2004). From a clinical perspective, adult attachment has been associated with domestic violence (Treboux et al., Reference Treboux, Crowell and Waters2004), psychiatric symptomatology (Vogel & Wei, Reference Vogel and Wei2005), and depression in non-clinical samples (Lopez, Mauricio, Gormley, Simko, & Berger, Reference Lopez, Mauricio, Gormley, Simko and Berger2001). In conclusion, attachment studies have measured romantic attachment in adults, particularly using the ECR, have been conducted in a variety of settings, and are addressing themes of great importance within psychology, and also to our understanding of romantic relationships.
Nevertheless, the majority of said studies have been carried out in North American and European populations (e.g., Alonso-Arbiol, Balluerka, & Shaver, Reference Alonso-Arbiol, Balluerka and Shaver2007; Lafontaine & Lussier, Reference Lafontaine and Lussier2003). Latin American countries are poorly represented within the body of research on adult attachment. This has made it difficult to generalize research findings to culturally distinct populations like ones in Latin America. One reason for the dearth of literature on romantic attachment in Chile and other Latin American countries is the lack of valid tests available to measure it and to conduct such studies (Martínez & Santelices, Reference Martínez and Santelices2005). Though the CAMIR adult attachment questionnaire was validated in Chile (Garrido, Santelices, Pierrehumbert, & Armijo, Reference Garrido, Santelices, Pierrehumbert and Armijo2009), it assesses adult attachment strategies in past and present relationships in general; it is not specific to romantic relationships. Thus, at a loss for an instrument to evaluate adult romantic attachment, it has been impossible to examine that variable in a Chilean population, nor to conduct cross-cultural comparisons to refine our understanding of romantic attachment and its expression in adults.
As Alonso-Arbiol and her colleagues (2007); (Shaver, Mikulincer, Alonso-Arbiol, & Lavy, Reference Shaver, Mikulincer, Alonso-Arbiol, Lavy, Erdman, Ng and Metzger2010) assert, to adapt the ECR, it would not only have to be translated. Researchers would need to determine its psychometric properties for the population and culture in which it will be used (Hambleton, Reference Hambleton1994). Toward that end, the present paper’s objective was to develop a valid, reliable version of the ECR so that studies of romantic attachment could be conducted in Chilean adults of many ages, not just young college students. This study’s second aim was to develop a short-form version of the questionnaire to be used in studies requiring an extensive battery of tests to be administered.
The objectives above were operationalized through a validation study based on exploratory factor analysis where in addition to analyzing the Chilean ECR’s psychometric properties and developing a short-form version of it, we have described how it was adapted into Chilean Spanish. This research was completed by applying confirmatory factor analysis to the short-form version created in the study.
Method
Participants
The sample was comprised of 773 adults (434 women and 339 men) from the Maule region of Chile. Of those, 555 belonged to 185 different families (father, mother, son, or daughter) that were taking part in a broader study of attachment’s transmission across generations. The remaining 218 participants (youths and adults) had no family ties to one another. Participants ranged in age from 18 to 78 years (M age = 38, SD = 14.45). In pursuit of our research objectives, independent factor analyses were conducted (exploratory and confirmatory), so the sample was divided randomly into two subsets. The first sub-sample was made up of 477 participants (56% women), while the remaining 296 formed the second sub-sample (56.4% women).
Regarding the sample’s distribution in terms of socioeconomic status, it was observed that 0.3% were lower class, 6.7% lower-middle class, 51.1% middle class, 29.8% upper-middle class, and 12.2% upper class. As for participants’ level of education, it is estimated that of the total sample, 295 were pursuing higher education at the time of assessment. The remaining participants are distributed as follows in terms of level of education: 0.6% no formal education, 4.2% incomplete elementary schooling, 6.6% completed elementary school, 25.1% incomplete secondary education, 14.2% completed secondary education, 12.6% incomplete vocational training, 12.0% completed vocational training or attended but did not complete college, 20.0% college graduates.
Instruments
Experiences in Close Relationships (ECR, Brennan et al., Reference Brennan, Clark, Shaver, Simpson and Rholes1998; the Spanish version validated in Spain was used, Alonso-Arbiol et al., Reference Alonso-Arbiol, Balluerka and Shaver2007). This self-report questionnaire assesses the dimensions of adult attachment. Participants respond to the original test’s 36 statements on a 7-point Likert-type scale from 1 (“not at all like me”) to 7 (“very much like me”).This generates two separate scores based on average scores on each dimension’s respective items, which correspond to the two dimensions of adult attachment: Avoidance (of intimacy) and Anxiety (about relationships). Low scores on the two scales indicate secure romantic attachment.
Due to differences between the Spanish spoken in Spain and Chile, items from the version validated in Spain were linguistically adapted (Alonso-Arbiol et al., Reference Alonso-Arbiol, Balluerka and Shaver2007) using an expert panel to ensure their validity. We chose to start with the Spanish version, more culturally similar than the original English version, to ease the translation (Balluerka, Gorostiaga, Alonso-Arbiol, & Haranburu, Reference Balluerka, Gorostiaga, Alonso-Arbiol and Haranburu2007). Following the suggestions of Balluerka and her colleagues, four psychologists familiar with attachment theory and/or romantic partner psychology independently assessed each original item, suggesting changes if the wording seemed inappropriate for a Chilean population. Later, the original and adapted items were evaluated by three of the four expert judges that participated in the first stage. They reviewed the proposed modifications’ fit to the meaning of the original items. Judges were asked if the items were suitable, somewhat suitable, or unsuitable. Heeding their suggestions, problematic items were adjusted and resubmitted to be newly assessed by the judges. Next, we determined interjudge agreement about the modifications, yielding a Kappa coefficient of interjudge reliability of 0.7. The main changes had to do with the word “intimacy” because in Chile, it is more closely associated with the sexual element of romantic relations. Therefore, a note was included at the beginning of the test introducing the term “intimacy” and explaining its meaning in the context of the questionnaire. The questionnaire’s internal consistency is high, yielding Cronbach’s alpha values of .83 (avoidance) and .84 (anxiety), very similar to those reported by Alonso-Arbiol and her colleagues (2007).
Sociodemographic Variables Questionnaire
Participants’ age, gender, and socioeconomic status (SES) were evaluated by means of a sociodemographic questionnaire. Socioeconomic status was determined in accordance with ADIMARK’s matrix (2004). This method is based on assessing two main variables. The first is the level of education attained by the household’s main financial provider (from no formal education up to post-graduate education). The second variable corresponds to goods whose presence or absence in the household discriminates among groups with different SESs. The two variables combine to form a socioeconomic status matrix. Initially, we assigned SESs values from 1 to 5 where 1 (lower class), 2 (lower-middle), 3 (middle), 4 (upper-middle), and 5 (upper). The variable was later dichotomized to create two groups with clearly distinct SESs. The values 1 and 2 were grouped together to generate a low SES sample (n = 54), while the values 4 and 5 were lumped together to constitute a high SES sample (n = 324). To gain discriminant power, the value 3 was not taken into account here, the objective being to forge two different groups in a sample whose middle socioeconomic stratum was overrepresented. Differences have been observed in the literature when it comes to the lower, or at-risk, strata in that they reportedly score higher on insecure attachment (e.g., van IJzendoorn & Bakermans-Kranenburg, Reference van IJzendoorn and Bakermans–Kranenburg2010).
Procedure
By contacting universities in the city of Talca, young college students were invited to take part in the study; they were also asked to invite their parents or some other adult to participate. In 185 cases, the young’s parents agreed to collaborate in the study. All participants took part in the research voluntarily, signed an informed consent form, and were assured of their anonymity and the confidentiality of their responses. The research team administered the battery of tests either at each participant’s college, or in his or her home. In the first case, families were reimbursed for any money spent commuting to the location where the tests were administered. Otherwise, participants received no incentive to participate. The questionnaires were completed individually (note that they were administered before an earthquake severely affected the region in February, 2010).
Data Analysis
Principal components analysis with varimax rotation was carried out. Subsequently, the reliability of the scales’ subdimensions was assessed using Cronbach’s alpha coefficient. Next, confirmatory factor analysis was performed on the second sample utilizing a parceling procedure. Concurrent validity was also analyzed using several variables. First, Pearson’s correlations were computed for the age variable. Second, parametric means comparison using Student’s t test was complemented by calculating effect size using Cohen’s d for the age variable. Third, means comparison using the Mann-Whitney U non-parametric test was supplemented by calculating effect size using etha square (η²) for the socioeconomic status variable.
Results
Principal components analysis with varimax rotation was applied to the original, 36-item version using the sub-sample of 447 participants. Prior to this analysis, the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was computed and Bartlett’s test of sphericity applied to determine whether or not factor analysis ought to be done. Both analyses yielded adequate results (KMO = .85) and in the case of Bartlett’s test of sphericity, statistically significant results, χ2(630) = 4,744.64 p < .0001.
In our analysis, the strategy was to extract factors based on the following criteria: (a) scree-plot, (b) the factors’ interpretability (having different semantic content), and (c) an item saturation pattern suggesting clearly defined, distinguishable factors (the majority of items had a loading of at least |.40| on its own factor and |.30| or less on the other). Based on criteria a and b, we concluded there was an underlying 2-factor structure like in the original English version (Brennan et al., Reference Brennan, Clark, Shaver, Simpson and Rholes1998) and the Spanish version (Alonso-Arbiol et al., Reference Alonso-Arbiol, Balluerka and Shaver2007). This solution explained 29.8% of variance. The third criterion also suggested the two-factor structure was satisfactory as well as parsimonious; only 6 out of 36 items did not meet the standards described above.
A series of decisions was made to create a short, more robust test to evaluate the Anxiety and Avoidance dimensions of adult attachment. First, the 6 items alluded to in the last paragraph were eliminated because they loaded on factors other than the dimensions they were designed to measure. Once that was done, item analyses were carried out (see Table 1). The objective was to eliminate any items with low item-total correlations (defined a priori under .40) or any item that, if eliminated, would improve the scale’s Cronbach’s alpha. Five new items were eliminated based on the first criterion, but none were discarded as a result of the second.
Table 1. Analysis of Items on the Chilean ECR by Dimension (n = 477)
After eliminating 11 items, principal components analysis with varimax rotation was carried out anew, producing a two-factor solution explaining 35.1% of total variance. The two dimensions’ lack of correlation (r = −.03) highlights the ECR’s distinctly orthogonal structure that also characterized the English (American) and Spanish (Spain) versions. The 25 final items’ factor loadings on each sub-dimension appear in Table 2. Internal consistency (Table 1) was computed by means of Cronbach’s alpha, which yielded good results (.83 in both cases). The final anxiety scale consisted of 11 items, and avoidance of 14.
Table 2. Factor Loadings of Items Retained after Principal Components Analysis
Note:
R = Reworded items. Items with factor loadings over .40 appear in bold.
Table 2. Factor Loadings of Remaining Items after Principal Components Analysis
Note:
R = Reworded items. Items with factor loadings over .40 appear in bold.
Confirmatory Factor Analysis
The two-factor structure was tested in the second sub-sample using confirmatory factor analysis and the method of maximum likelihood estimation. Following the procedure that Alonso-Arbiol, Balluerka, Shaver, and Gillath (Reference Alonso-Arbiol, Balluerka, Shaver and Gillath2008) employed to determine the factor structure of the original ECR in English and the Spanish version created in Spain, here too we employed a parceling technique. To reduce the number of parameters to be estimated, making the model easier to identify, the B-ECR’s 25 items were grouped into 9 parcels (for an in-depth explanation of this procedure, see Little, Cunningham, Shahar, & Widaman, Reference Little, Cunningham, Shahar and Widaman2002).
According to the criteria proposed by Hoyle (Reference Hoyle1995) and Byrne (Reference Byrne2010), we report that all the goodness of fit statistics used were found to be adequate (χ2/df: 3.15, AGFI: .90, TLI: .91, CFI: .95, RMSEA: .08). This supports the two-factor model.
Analysis of Concurrent Validity
The test’s concurrent validity was examined in terms of its dimensions’ associations with age and gender variables in the total sample, and with SES in the sub-sample. In the literature, the dimensions of attachment seem to be unrelated to gender (e.g., Brennan et al., Reference Brennan, Clark, Shaver, Simpson and Rholes1998; Alonso-Arbiol et al., Reference Alonso-Arbiol, Balluerka and Shaver2007) and age (e.g., Alonso-Arbiol, Shaver, & Yárnoz, Reference Alonso-Arbiol, Shaver and Yárnoz2002). In our study, correlations between age and avoidance (r = .05) and between age and anxiety (r = .03) support that non-association. As for the means comparison between men and women on the two dimensions of attachment, no differences were observed in this study, neither in the case of avoidance, t(771) = −0.87, p = .39, M = 2.59, SD = 1.00 in women and M = 2.66, SD = 0.93 in men, Cohen’s d = .06); nor anxiety, t(771) = .01, p = .99, M = 4.25, SD = 1.12 in women and M = 4.26, SD = 1.12 in men, Cohen’s d = .001).
With respect to socioeconomic status (SES), in keeping with our expectations (e.g., van IJzendoorn & Bakermans-Kranenburg, Reference van IJzendoorn and Bakermans–Kranenburg2010), anxiety discriminated between the low and high SES sub-samples, but the magnitude of that difference was small (η² = .04). Meanwhile, the low and high SES sub-samples’ avoidance levels did not differ. These results are displayed in Table 3.
Table 3. SES Differences in the Dimensions of Attachment
Note:
SES = Socioeconomic Status.
Chilean B-ECR Standards
So as to provide norms for interpreting scores on the Chilean B-ECR in related, future studies, we computed percentages in the total sample of 772 participants (see Table 4). This is crucial because total scores are computed using averages of item values, not the sum of their values, so one must retain both decimal points when using this test in assessment. Since gender differences were not observed, the same standards apply to men and women alike.
Table 4. Chilean Standards for Dimensions of Attachment
Discussion
The present study was conducted with the objective to adapt and examine the psychometric properties of a short-form version of the ECR to assess adult romantic attachment in a Chilean population. Applying principal components analysis to an earlier, more extensive version, as well as confirmatory factor analysis to a 25-item, short-form version clearly revealed a two-factor structure. The internal consistency of the ECR’s two sub-scales was also found to be optimal. Moreover, additional evidence for its concurrent validity in terms of the variables age, gender, and socioeconomic status reinforces that the Chilean short-form version of the ECR (B-ECR) is a valid, reliable tool to assess attachment in Chilean adult populations of various ages.
The observed lack of gender differences in the dimensions of adult attachment is consistent with the literature on this subject (Brennan et al., Reference Brennan, Clark, Shaver, Simpson and Rholes1998). No statistically significant differences have been reported in the disparate samples in which this instrument has been applied, not even adult samples of different ages (Alonso-Arbiol et al., Reference Alonso-Arbiol, Shaver and Yárnoz2002). Similarly, the finding that scores on the anxiety and avoidance dimensions do not vary as a function of gender are consistent with earlier findings about romantic attachment (Lafontanie & Lussier, Reference Lafontaine and Lussier2003; Olsson, Sorebo, & Dahl, Reference Olsson, Sorebo and Dahl2010; Tsagarakis, Kafetsios, & Stalikas, Reference Tsagarakis, Kafetsios and Stalikas2007). In that vein, the findings favor the hypothesis that the gender variable does not affect the quality of attachment in adults (Bakermans-Kranenburg & van IJzendoorn, Reference Bakermans-Kranenburg and van IJzendoorn2009; Mikulincer & Shaver, Reference Mikulincer and Shaver2007; van IJzendoorn & Bakermans-Kranenburg, Reference van IJzendoorn and Bakermans–Kranenburg2010). Conversely, a recent meta-analysis reported significant differences between men and women in terms of romantic attachment (Del Giudice, Reference Del Giudice2011). Its results, however, should be taken with a grain of salt considering that, as the author himself suggests, no research samples whatsoever were taken from Latin American populations. Furthermore, Del Guidice’s meta-analysis was carried out using different instruments. That being said, a cross-cultural study of gender differences on the ECR (Alonso-Arbiol et al., Reference Alonso-Arbiol, van de Vijver, Shaver, Azaiza, Brassard, Çelenk and Spencer2010) did report gender differences occurring in highly masculine countries (i.e. Italy, Japan), that is, according to Hofstede’s conception of masculine and feminine (1998, 2001). According to his classification, Chile is considered a feminine country, so one would not expect gender differences in the dimensions of attachment. This notion was supported by data from the Chilean sub-sample in the study by Alonso-Arbiol and her colleagues (2010).
Regarding the association between low SES and anxiety, and the non-association between SES and the avoidance dimension, the results partially confirmed our expectations. However, the literature that directly links socioeconomic status to attachment usually stems from non-romantic conceptualizations of adult attachment. For example, van IJzendoorn, and Bakermans-Kranenburg (Reference van IJzendoorn and Bakermans–Kranenburg2010), who evaluated adult attachment using the Adult Attachment Interview (AAI; George, Kaplan & Main, Reference George, Kaplan and Main1985), observed an over-representation of dismissive and preoccupied attachment (similar to anxiety) in a sample of men with low SES. These results assert that socioeconomic status, particularly poverty and the difficulties associated with it, is a variable that can affect security of attachment in adults. The differential datum reported in our study, that low SES is only linked to anxiety (and not avoidance), might tentatively lead us to argue that an unfavorable economic situation specifically influences the dimension of attachment related to the need to find a partner. From a material standpoint, increased risk of losing one’s partner would also entail losing the complementary resources a spouse can provide. Therefore, it stands to reason that one might adopt a more anxious attachment style where each member is overly vigilant of the other’s presence and interest in continuing the relationship. From a more psychological perspective, and bearing in mind the association between experiencing negative or unfavorable life events and higher levels of attachment anxiety (Drake, Sheffield, & Shingler, Reference Drake, Sheffield and Shingler2011), the support of one’s partner (attachment figure) is key to decreasing that kind of anxiety. If more situations threaten one’s security of attachment, perhaps the attachment system is continuously activated. If that were the case, a common strategy might be to rely on one’s partner being constantly present to alleviate that anxiety.
This study makes one unequivocal contribution to the literature, a short-form test. Increasingly many authors advocate for creating short-form surveys. They are advantageous in that they can be more rapidly administered, and sample participants prefer them (e.g., Gallarin & Alonso-Arbiol, in press; Gorostiaga, Balluerka, Alonso-Arbiol, & Haranburu, Reference Gorostiaga, Balluerka, Alonso-Arbiol and Haranburu2011).
On a different note, this cultural adaptation deserves special mention. Balluerka and her colleagues (2007) argue the importance of cultural and linguistic aspects in adaptation, even when the same language is used. This aspect was previously emphasized in attachment research using the ECR (Alonso-Arbiol et al., Reference Alonso-Arbiol, Balluerka and Shaver2007; Shaver et al., Reference Shaver, Mikulincer, Alonso-Arbiol, Lavy, Erdman, Ng and Metzger2010). Other Chilean authors have suggested the need to provide national standards for questionnaires created or validated in Spain to reflect the cultural specificity of the population in which the instrument will be used (Fernández, Dufey, & Kramp, Reference Fernández, Dufey and Kramp2011; Guerra Vio, Castro Arancibia, & Vargas Castro, Reference Guerra Vio, Castro Arancibia and Vargas Castro2011). The short-form ECR devised here is useful due to its brevity and its reliability. This study’s results can be considered a starting point to study several variables associated with adult attachment in the Chilean context. In addition to its benefits to advancing research, developing this short-form version and reporting Chilean standards will be of tremendous use to the multitude of professionals working in applied psychology.
Nevertheless, the present study is not without limitations that should be kept in mind and overcome in future research. Among its limitations is the sample utilized; it is not representative of the Chilean population because a considerable percentage of families (fathers-mothers and sons or daughters) of young college students were included. This created a bias toward over-representation of middle-class participants. Notwithstanding, this limitation, this sample was more representative than others made up entirely of college students (e.g., Alonso-Arbiol et al., Reference Alonso-Arbiol, Shaver and Yárnoz2002; Brennan et al., Reference Brennan, Clark, Shaver, Simpson and Rholes1998; Wongpakaran, Wongpakaran, & Wedding, Reference Wongpakaran, Wongpakaran and Wedding2011) in which adults of other ages have been included. The sample’s SES posed an additional limitation in that the poorest sectors of the population were very under-represented. Conversely, the middle-class was very well represented. This could be explained by how the sample was gathered. Participants were contacted through universities, to which only a portion of the population has access.
Future research could develop a short-form version of the ECR in non-Chilean, Spanish-speaking populations and test its psychometric properties. That way, a cross-cultural study of adult attachment could be conducted. In addition, future research should more deeply analyze the instrument’s external validity and examine the relationship between adult romantic attachment, as evaluated by the B-ECR, and other psychological and behavioral variables that the literature shows may correlate with romantic attachment.
