16
European Journal of Psychological Assessment Selecting the Best Items for a Short-Form of the Experiences in Close Relationships Questionnaire Marie-France Lafontaine, Audrey Brassard, Yvan Lussier, Pierre Valois, Philip R. Shaver, and Susan M. Johnson Online First Publication, February 27, 2015. http://dx.doi.org/10.1027/1015-5759/a000243 CITATION Lafontaine, M.-F., Brassard, A., Lussier, Y., Valois, P., Shaver, P. R., & Johnson, S. M. (2015, February 27). Selecting the Best Items for a Short-Form of the Experiences in Close Relationships Questionnaire. European Journal of Psychological Assessment. Advance online publication. http://dx.doi.org/10.1027/1015-5759/a000243

Selecting the Best Items for a Short-Form of the Experiences in Close Relationships Questionnaire

Embed Size (px)

Citation preview

European Journal of PsychologicalAssessment

Selecting the Best Items for a Short-Form of theExperiences in Close Relationships QuestionnaireMarie-France Lafontaine, Audrey Brassard, Yvan Lussier, Pierre Valois, Philip R. Shaver, andSusan M. JohnsonOnline First Publication, February 27, 2015. http://dx.doi.org/10.1027/1015-5759/a000243

CITATIONLafontaine, M.-F., Brassard, A., Lussier, Y., Valois, P., Shaver, P. R., & Johnson, S. M. (2015,February 27). Selecting the Best Items for a Short-Form of the Experiences in CloseRelationships Questionnaire. European Journal of Psychological Assessment. Advance onlinepublication. http://dx.doi.org/10.1027/1015-5759/a000243

Original Article

Selecting the Best Items for aShort-Form of the Experiences in

Close Relationships QuestionnaireMarie-France Lafontaine,1 Audrey Brassard,2 Yvan Lussier,3 Pierre Valois,4

Philip R. Shaver,5 and Susan M. Johnson6

1Department of Psychology, University of Ottawa, Canada, 2Université de Sherbrooke, Canada,3Université du Québec à Trois-Rivières, Canada, 4Université Laval, Canada, 5University of California,

Davis, CA, USA, 6University of Ottawa, Ottawa Couple and Family Institute, Canada

Abstract. Five studies were conducted to develop a short form of the Experiences in Close Relationships (ECR) questionnaire with optimalpsychometric properties. Study 1 involved Item Response Theory (IRT) analyses of the responses of 2,066 adults, resulting in a 12-item form ofthe ECR containing the most discriminating items. The psychometric properties of the ECR-12 were further demonstrated in two longitudinalstudies of community samples of couples (Studies 2 and 3), in a sample of individuals in same-sex relationships (Study 4), and with couplesseeking therapy (Study 5). The psychometric properties of the ECR-12 are as good as those of the original ECR and superior to those of anexisting short form. The ECR-12 can confidently be used by researchers and mental health practitioners when a short measure of attachmentanxiety and avoidance is required.

Keywords: attachment, self-report measure, Experiences in Close Relationships, psychometrics, Item Response Theory

Interest in adult attachment among researchers and clinicalpractitioners has increased over the past 25 years, spurringefforts to create and validate self-report measures of attach-ment in romantic, or couple, relationships (see Mikulincer& Shaver, 2007, for a review). One of the most frequentlyused attachment questionnaires is the Experiences in CloseRelationships (ECR) questionnaire, developed by Brennan,Clark, and Shaver (1998). This 36-item questionnaire is asynthesis of previous measures created by many differentinvestigators and is based on factor analyses of 323 itemsfrom such measures. Translated into many languages, theECR is now considered a reference instrument. It assessesanxiety concerning rejection or abandonment and avoid-ance of intimacy and interdependence.

The anxiety dimension is associated with viewing one-self as needing others but being vulnerable to rejection(Bartholomew & Horowitz, 1991; Collins & Allard,2001). The avoidance dimension is associated with a work-ing model of others (i.e., relationship partners) as overlyneedy and dependently intrusive, and with a strong desireto remain independent and self-reliant (Bartholomew &Horowitz, 1991; Collins & Allard, 2001). These two dimen-sions are solidly grounded in attachment theory (Bowlby,

1969/1982) and research, beginning with the pioneeringwork of Ainsworth, Blehar, Waters, and Wall (1978) oninfant-mother attachment.

The two-dimensional factor structure of the ECR hasbeen observed in samples of men and women (Lafontaine& Lussier, 2003), and in homosexual and heterosexual,clinical and nonclinical adult populations (Alonso-Arbiol,Balluerka, & Shaver, 2007; Lafontaine & Lussier, 2003;Matte, Lemieux, & Lafontaine, 2009; Picardi, Bitetti,Puddu, & Pasquini, 2000). This structure is also evidentin many translations of the ECR into languages other thanEnglish, including Chinese (Mallinckrodt & Wang, 2004),Dutch (Conradi, Gerlsma, van Duijn, & de Jonge, 2006),French (Lafontaine & Lussier, 2003), Hebrew (Mikulincer& Florian, 2000), Italian (Picardi et al., 2000), Japanese(Nakao & Kato, 2004), and Spanish (Alonso-Arbiol et al.,2007). The ECR has been used worldwide because of itshigh reliability and validity, demonstrated in many correla-tional and experimental studies; for example: (a) convergentvalidity with measures of relationship satisfaction, psycho-logical distress, fear of intimacy, romantic dependence,accommodation strategies, and self-esteem, among otherconstructs; (b) high test-retest reliability (ranging from .50

� 2015 Hogrefe Publishing European Journal of Psychological Assessment 2015DOI: 10.1027/1015-5759/a000243

to .75 depending on the nature of the study and the timespan being assessed); and (c) high internal consistency(for both subscales, anxiety and avoidance, near or above.90; e.g., Alonso-Arbiol et al., 2007; Brennan et al., 1998;Conradi et al., 2006; Lafontaine & Lussier, 2003;Mallinckrodt & Wang, 2004; Nakao & Kato, 2004).Because the two ECR scales are fairly long (18 items each),there is a strong desire among researchers for a shorter ver-sion of the ECR based on the best discriminating items andretaining high reliability. Using a restricted pool of itemshas the advantage of reducing participants’ fatigue, frustra-tion, and boredom (Robins, Hendin, & Trzesniewski, 2001).Furthermore, a brief ECR would be very useful in large-sample studies or as part of a thorough psychologicalassessment. The primary purpose of our five studies wasto develop a short version of the ECR and to validate itin five samples, including couples from the general com-munity, individuals in same-sex relationships, and couplesseeking therapy for relationship difficulties. Based on thischoice, we used relationship satisfaction and psychologicaldistress as validity criteria. The selected pool of items hasto be small enough to allow a rapid assessment of attach-ment dimensions but large enough to ensure accurateparameter estimates and good reliability (Marsh, Hau,Balla, & Grayson, 1998). Based on Marsh et al.’s (1998)recommendation that a factor should include a minimumof four items, our goal was to select a minimum of eightitems (i.e., four items for each ECR dimension). Other stud-ies have followed this recommendation in developing short-form questionnaires (e.g., Poitras, Guay, & Ratelle, 2012).

Previous Modifications of the ECR

To our knowledge, only one short version of the ECR hasbeen published, a 12-item questionnaire created by Wei,Russell, Mallinckrodt, and Vogel (2007), called theECR-S. They conducted separate exploratory factor analy-ses on items from the anxiety and avoidance subscales.They identified redundant items based on inter-item corre-lations and similar wording, and retained only items withthe highest factor loadings. This logical and statisticalapproach resulted in a six-item avoidance subscale (includ-ing the original ECR items numbered 11, 13, 17, 27, 33,and 35) and a six-item anxiety subscale (items 6, 16, 18,22, 26, and 32). Wei et al. (2007) also performed confirma-tory factor analyses (on the ECR-S and the ECR) to test dif-ferent models using college and undergraduate samples,including the sample on which the item selection wasbased. The fit to the models was comparable for the shortversion and the original version, but this was the case onlyafter two latent variables, representing positively and nega-tively phrased items, were statistically controlled.

Five studies assessed the validity and reliability of theECR-S, and all of them yielded acceptable internal consis-tency for both short subscales. The same five studies eval-uated the construct validity of the ECR-S using correlationswith relevant constructs (e.g., depression, anxiety, andsocial desirability). Analyses confirmed that test-retest

reliability for the ECR-S was acceptable over both a 3-weekinterval (Study 6) and a 1-month interval (Study 4).

Wei and colleagues (2007) demonstrated that it is possi-ble to reduce the number of items on the ECR while pre-serving its validity and reliability, but several limitationsof their work justify the creation of an alternative shortform. First, the generalizability of their validation study islimited by its reliance on homogeneous samples of NorthAmerican undergraduate students. Second, the factor struc-ture of the ECR-S was acceptable only after controlling fortwo additional latent variables accounting for response sets(which was not the case with the original ECR). Third, Weiet al. (2007) used Classical Test Theory (CTT) to selectitems to be removed or retained. However, many research-ers have argued that CTT has a number of shortcomings(e.g., Embretson & Reise, 2000; Thissen & Steinberg,1997). For instance, because test analysis in CTT occursat the level of the test rather than the item, it provides littleinformation concerning how a specific item functions inde-pendently of the other items (Stroud, McKnight, & Jensen,2004). Another limitation of CTT is that its estimates ofitem parameters are dependent on the particular samplefrom which they were derived. Finally, CTT assumes anaverage standard error of measurement for all examinees.

These shortcomings have motivated psychometriciansto seek alternative theories and models to measure mentalprocesses. According to Nunnally and Bernstein (1994),Item Response Theory (IRT) methods are central in modernpsychometrics and are now considered a good alternative todevelop and validate psychosocial scales. Although, CTT-and IRT-based scales can be highly comparable (e.g.,MacDonald & Paunonen, 2002), IRT allows selection ofitems providing maximum information within a specifictrait range along the underlying latent dimension (e.g., anx-iety or avoidance) and the creation of shorter tests that canbe even more reliable than corresponding longer tests(Embretson & Reise, 2000).

Fraley, Waller, and Brennan (2000) used IRT analyseson Brennan et al.’s (1998) original pool of 323 attachmentitems to create the ECR-R. To improve discrimination atthe secure ends of the two dimensions, Fraley et al. replaced18 items in the original ECR while retaining the other 18, 6from the original anxiety subscale (items 4, 6, 10, 16, 22,and 26) and 12 from the avoidance subscale (items 1, 3,7, 9, 13, 15, 19, 21, 23, 25, 27, and 29). Eighteen new itemsfrom the original 323 were chosen to replace the 18 thatwere eliminated, making the ECR-R another 36-item mea-sure. Several studies have provided support for the validityand reliability of the ECR-R (e.g., Fairchild & Finney,2006; Sibley, Fischer, & Liu, 2005). Despite this support,Mikulincer and Shaver (2007) have continued to use theoriginal ECR, arguing that: (a) the wording of some ofthe new items on the ECR-R is problematic; (b) there is aslightly higher correlation between the two dimensions(anxiety and avoidance) of the ECR-R than is characteristicof the original ECR, perhaps reducing discriminability ofthe two constructs; (c) the ECR-R does not yield a signifi-cant increase in validity over the ECR, and (d) the ECR andthe ECR-R share 18 items, which results in the correlationbetween the two versions of the anxiety and avoidance

2 M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12

European Journal of Psychological Assessment 2015 � 2015 Hogrefe Publishing

subscales being approximately .95. For these reasons, wedecided to construct our short scales using the originalECR items.

The Present Research

The goal of the first study reported here was to develop ashort version of the ECR. IRT analyses were used to exam-ine item performance in the original ECR, as a basis forselecting items for a shorter measure (Study 1). Four inde-pendent studies were then conducted to examine whetherthis new short measure presented with superior psychomet-ric properties than the ECR-S across various participantsamples: two community samples of French-Canadian cou-ples (Study 2) and English-speaking couples (Study 3), acommunity sample of English-speaking gay, lesbian, andbisexual adults (Study 4), and a clinical sample of Canadiancouples seeking couple therapy (Study 5). Gender-relateddifferences were tested based on the results of a meta-analysis showing that women reported higher anxiety andlower avoidance than men, particularly in communitysamples (Del Giudice, 2011).

Study 1

The objectives of Study 1 were: (1) to select ECR items fora new short form, based on two IRT criteria; and (2) to per-form differential item functioning (DIF) analyses (Ramsay,2000) on the selected items to determine whether the rangeof answers on a given item was biased by gender (i.e.,whether men and women used a different range ofresponses on the item).

Method

Participants and Procedure

Participants were 2,066 French-Canadian college students,1,301 women (63.3%), and 755 men (36.7%). The meanage of the sample was 20.69 years (SD = 3.18), with anaverage of 13 years of education. The majority of the par-ticipants (62.3%) were involved in a romantic relationshipat the time of the study, and 72 participants (3.5%) had chil-dren. Research assistants recruited participants in variouscolleges in the province of Québec, and the participantswere provided questionnaires and were instructed to com-plete them in class (56%) or at home (44%) and to returnthem in the preaddressed envelope provided.

Measures

The attachment dimensions – anxiety and avoidance – wereassessed with a French-language version of the ECR

(Lafontaine & Lussier, 2003). Participants indicated theirlevel of agreement with each statement using a seven-pointrating scale ranging from 1 (= disagree strongly) to7 (= agree strongly), based on their general orientation inclose relationships. Items on both scales were averaged(following appropriate reversals of negatively wordeditems) to produce two scores for each participant; higherscores indicated greater anxiety and avoidance.

Data Analysis Strategy

To determine which items to include in the ECR short form,we used two criteria based on IRT (Reeve & Fayers, 2005).Several models of IRT can be applied to rating data but,given the use of Likert scales, the Graded-Response Model(GRM; Samejima,1969, 1997) was deemed most appropri-ate (Embretson & Reise, 2000). The primary objective ofIRT methods is to specify the associations between itemresponses and the latent trait (h) posited to be measuredby the test or questionnaire. An Item Characteristic Curve(ICC) graphically represents the way in which the probabil-ity of a response varies with the level of the underlying trait.In the graded response model, if the probability of endors-ing an item or option increases as a function of, forinstance, anxiety, the item is effective. This first criterionis called ‘‘discrimination.’’

In order to compute the parameter estimates under theGRM, the EIRT program (Excel Item Response Theoryassistant; Valois, Houssemand, Germain, & Abdous,2011) was used in the current study. This program estimatesthe ability of items and scales to detect differences betweenindividuals across a latent trait. Ideally, the differencesbetween two individuals should be easily detected, regard-less of whether they are in the low or high regions of thelatent trait. The slope of the ICC (ai; item discriminability)indicates the extent to which a change in item score corre-sponds to changes in the level of latent trait (Baker, 2001).

We examined results of item analyses for each of theoriginal ECR’s 18 avoidance items and its 18 anxiety items.Our goal was to select the items that best differentiated par-ticipants in terms of avoidance and anxiety scores along theentire span of scores. This was done by choosing the itemsthat yielded the steepest (most steeply sloped) item charac-teristic curves, with steepness indicating how well theitem distinguishes participants with different trait levels.To assess the precision of these slopes, we created 1,000nonparametric bootstrap samples of 1,033 participants (arandom half of the total sample in each bootstrap sample),a procedure designed to empirically estimate the variabilityof a statistic (the slope of the ICC) whose theoretical distri-bution is unknown. We ‘‘resampled’’ the data from the ori-ginal sample with replacement to generate an empiricalestimate of the entire sampling distribution of the slopefor each item (Mooney & Duval, 1993). The analysis wasperformed on each bootstrap sample, which gave 1,000slope values and standard error estimates for every item.All of these values were then ranked, averaged, and usedto construct 99% confidence intervals, allowing us to iden-tify the most discriminating items.

M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12 3

� 2015 Hogrefe Publishing European Journal of Psychological Assessment 2015

After analyzing the degree of discrimination, we exam-ined the item difficulty for all items with the bij parametersand their standard errors. The bij parameters represent agiven avoidance or anxiety level as a function of the max-imum probability of choosing each response option. For thechoice of response 2 on item 2, for example, the bij param-eter b22 is the place on the latent trait (or theta scale) of anx-iety where the probability of choosing option 2 is maximal.Good items have bij parameters that match their respectiveresponse levels and have minimal error (Poitras et al.,2012). Based on Baker (2001), we wanted to keep itemswith bij parameters having an adequate location on the larg-est possible scale continuum (from �3 to 3) and the loweststandard errors. Based on these two IRT criteria (item dis-crimination and difficulty), the bivariate correlationsbetween each possible short scale (from 2- to 18-item anx-iety and avoidance subscales) and the original ECR sub-scales, and the distribution of their standard errors, wedetermined the best number of items for the new shortform. We then conducted differential item functioning(DIF; Ramsay, 2000) analyses on the selected items toexamine the presence of gender bias.

Results of Study 1

IRT Analysis for the Avoidance Items

Following the estimation of the slope of each item’s itemcharacteristic curve, using the 1,000 bootstrap samples,we ranked the 18 avoidance items based on their slopesusing EIRT software (Valois et al., 2011). This allowed usto identify how many times an item had the steepest slope(i.e., ranked first) over the 1,000 bootstraps, how many

times an item had the second steepest slope, etc. Table 1shows the ranks of the 18 avoidance items. For example,item 29, which possessed the highest mean slope, wasranked first on 427 occasions, second on 238 occasions,third on 193 occasions, etc.

Table 1 suggests that there were four groups of items,based on their ranks: Items 29, 27, 25, 31, 9, and 15(group 1: with item slopes found mostly in the first sixthranks) had the best rankings, followed by items 23, 7, 13,and 19 (group 2: with item slopes found mostly betweenranks 7 and 10, inclusive); items 5, 11, 1, 21, and 17 (group3: with item slopes found mostly between ranks 11 and 15,inclusive); and items 33, 3, and 35 (group 4: with itemslopes found mostly between ranks 16 and 18, inclusive).Among the six items with the steepest slopes (i.e.,group 1), none appeared among the last five ranks overthe 1,000 bootstraps. Items 29, 27, and 25 were classifiedmore than 900 times in the top five ranks, followed byitem 9 (721 times), item 31 (671 times), and item 15(482 times). In contrast, items from the three other groupswere never or rarely among the top six ranks.

Table 2 presents, in decreasing order, the means andconfidence intervals for slopes of the item characteristiccurves (ICC) for each of the original 18 avoidance items.The averaged item ICC slopes were all quite high, althoughthe six items from group 1 appeared to be the best for ashort avoidance subscale. The estimation of those six items’slopes yielded the lowest standard errors, indicating greaterprecision. Item difficulty parameters for the 18 avoidanceitems, also shown in Table 2, revealed that these six itemshad adequate locations along most of the avoidance thetascale continuum (from �.96 to 2.86) and had low standarderrors (< .18). It is worth mentioning that almost all avoid-ance items performed relatively poorly on the lower portion

Table 1. Rank of items assessing avoidance based on estimated slopes using 1,000 bootstrap samples

Item

Rank 29 27 25 31 9 15 23 7 13 19 5 11 1 21 17 33 3 35

18 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 7 99317 0 0 0 0 0 0 0 0 1 0 2 2 1 1 7 320 659 716 0 0 0 0 0 0 0 3 10 0 26 32 17 33 82 503 294 015 0 0 0 0 0 0 3 25 39 22 83 94 99 197 311 90 37 014 0 0 0 0 0 0 18 28 60 48 107 113 126 234 230 34 2 013 0 0 0 5 0 1 31 47 71 88 124 130 173 158 149 22 1 012 0 0 0 3 0 3 63 71 93 121 152 122 142 130 85 15 0 011 0 1 0 4 0 8 94 86 91 104 156 140 132 99 75 10 0 010 0 4 1 18 2 11 100 106 138 160 137 126 109 56 28 4 0 09 2 3 5 21 10 23 166 116 135 171 87 89 96 53 21 2 0 08 1 9 4 36 13 60 195 166 127 142 59 80 68 32 8 0 0 07 1 19 19 67 53 95 217 162 122 119 39 41 35 7 4 0 0 06 11 38 29 175 201 317 61 58 60 19 11 18 2 0 0 0 0 05 38 69 81 196 267 238 28 48 21 5 7 2 0 0 0 0 0 04 89 117 158 218 216 141 15 31 8 1 2 4 0 0 0 0 0 03 193 223 222 141 112 68 4 19 11 0 3 4 0 0 0 0 0 02 238 276 267 79 82 28 4 13 6 0 5 2 0 0 0 0 0 01 427 241 214 37 44 7 1 21 7 0 0 1 0 0 0 0 0 0

Note. Items in bold are the avoidance items retained in the ECR-S (Wei et al., 2007).

4 M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12

European Journal of Psychological Assessment 2015 � 2015 Hogrefe Publishing

Table 2. Slopes’ means, standard errors, and 99% confidence intervals, and item difficulty coefficients for the avoidance subscale items

Item a (SE) 99% CI b1 (SE) b2 (SE) b3 (SE) b4 (SE) b5 (SE) b6 (SE) b7 (SE)

29. I feel comfortable depending onromantic partners.

5.47 (.08) [5.25–5.66] �.77 (.05) �.25 (.03) .63 (.04) 1.28 (.07) 1.75 (.09) 2.26 (.12) 2.58 (.15)

27. I usually discuss my problemsand concerns with my partner.

5.44 (.10) [5.14–5.66] �.82 (.05) �.28 (.03) .64 (.04) 1.26 (.07) 1.76 (.09) 2.44 (.13) 2.86 (.17)

25. I tell my partner just abouteverything.

5.43 (.10) [5.18–5.65] �.96 (.06) �.41 (.03) .49 (.03) 1.09 (.06) 1.60 (.08) 2.33 (.12) 2.81 (.16)

31. I don’t mind asking romanticpartners for comfort, advice, orhelp.

5.34 (.11) [5.03–5.61] �.70 (.05) �.17 (.03) .77 (.04) 1.44 (.08) 1.95 (.10) 2.51 (.14) 2.82 (.17)

9. I don’t feel comfortable openingup to romantic partners.

5.34 (.09) [5.05–5.56] �.60 (.05) �.13 (.03) .57 (.04) 1.05 (.06) 1.60 (.09) 2.23 (.12) 2.57 (.14)

15. I feel comfortable sharing myprivate thoughts and feelingswith my partner.

5.30 (.09) [5.06–5.50] �.84 (.06) �.31 (.04) .53 (.04) 1.06 (.06) 1.48 (.08) 2.09 (.11) 2.49 (.14)

23. I prefer not to be too close toromantic partners.

5.12 (.13) [4.78–5.38] �.49 (.04) �.01 (.03) .72 (.04) 1.28 (.07) 1.90 (.10) 2.67 (.15) 3.12 (.20)

7. I get uncomfortable when aromantic partner wants to be veryclose.

5.11 (.19) [4.57–5.55] .15 (.03) .57 (.04) 1.14 (.07) 1.50 (.09) 2.01 (.12) 2.59 (.16) 2.87 (.19)

13. I am nervous when partners wantto get too close to me.

5.07 (.18) [4.59–5.46] .06 (.03) .46 (.04) 1.04 (.06) 1.41 (.08) 1.95 (.11) 2.73 (.17) 3.18 (.23)

19. I find it relatively easy to getclose to my partner.

5.05 (.12) [4.73–5.34] �.90 (.06) �.28 (.03) .75 (.04) 1.47 (.08) 2.08 (.11) 2.80 (.16) 3.21 (.20)

5. Just when my partner starts to getclose to me I find myself pullingaway.

4.98 (.18) [4.49–5.38] �.12 (.03) .38 (.03) 1.08 (.07) 1.50 (.09) 2.03 (.12) 2.65 (.16) 2.95 (.18)

11. I want to get close to my partner,but I keep pulling back.

4.98 (.18) [4.50–5.41] �.18 (.04) .25 (.03) .86 (.06) 1.33 (.08) 2.01 (.11) 2.67 (.16) 2.93 (.18)

1. I prefer not to show a partner how Ifeel deep down.

4.96 (.13) [4.63–5.27] �.81 (.06) �.26 (.04) .54 (.04) 1.08 (.06) 1.78 (.09) 2.64 (.14) 3.08 (.18)

21. I find it difficult to allow myselfto depend on romantic partners.

4.90 (.13) [4.56–5.21] �.90 (.06) �.35 (.04) .46 (.04) 1.14 (.06) 1.91 (.10) 2.70 (.15) 3.15 (.18)

17. I try to avoid getting too close tomy partner.

4.85 (.15) [4.43–5.20] �.51 (.04) �.06 (.03) .63 (.04) 1.19 (.07) 1.99 (.11) 2.89 (.16) 3.32 (.21)

33. It helps to turn to my romanticpartner in times of need.

4.58 (.17) [4.17–5.00] �.82 (.05) �.23 (.03) .84 (.05) 1.73 (.09) 2.49 (.13) 3.28 (.20) 3.73 (.28)

3. I am very comfortable being closeto romantic partners.

4.48 (.18) [4.00–4.87] �.14 (.04) .54 (.05) 1.47 (.09) 1.94 (.12) 2.29 (.14) 2.64 (.16) 2.83 (.18)

35. I turn to my partner for manythings, including comfort andreassurance.

3.85 (.17) [3.44–4.27] �1.57 (.11) �.90 (.07) .47 (.05) 1.75 (.10) 2.73 (.15) 3.86 (.23) 4.61 (.31)

Note. Items in bold are the avoidance items retained for the ECR-12. Italicized items are the avoidance items retained in the ECR-S (Wei et al., 2007).

M.-F.

Lafontaine

etal.:

Psychom

etricsP

ropertiesof

theE

CR

-125

�2015

Hogrefe

Publishing

European

Journalof

Psychological

Assessm

ent2015

of the avoidance scale continuum. Only item 35 had an ade-quate location on a lower portion of the theta continuum,but it had a high standard error and the lowest ICC of allthe avoidance items.

To further validate our choice of the best six items for abrief avoidance subscale, we compared versions of theavoidance subscale of different lengths with respect to theirdistributions of standard errors across points on the latentavoidance continuum. The results indicate that the gain ofadding more items decreases after six items, supportingour choice of a six-item avoidance subscale (Figure 1).We then compared the bivariate correlations between theoriginal ECR avoidance subscale and brief versions of dif-ferent lengths based on their ranks in Table 2. Results indi-cated that the correlation between the original ECRavoidance subscale and the one-item version is .722. Thiscorrelation reached .803 with the two-item version, .838with the three-item version, .864 with the four-item ver-sions, .889 with the five-item version, and .921 with thesix-item version, which is very good for an abridged scale.In sum, 6 of the original 18 items from the ECR avoidancesubscale displayed characteristics that made them suitablefor a briefer subscale (items 9, 15, 25, 27, 29, and 31). Inter-estingly, when using IRT, only item 27 was the same asthose retained in Wei et al.’s (2007) short version of theavoidance subscale.

IRT Analysis of the Attachment Anxiety Items

We used the same procedures described above to identify thebest items for a short-form anxiety subscale. Table 3 showsthe rankings of the 18 anxiety items based on 1,000 estimatedslopes. There were, again, four groups of items based on IRTcriteria: items 6, 8, 2, 14, 18, and 24 (group 1) have the

highest rankings, followed by items 20, 4, 30, and 26(group 2); items 22, 36, 32, and 10 (group 3); and items28, 12, 34, and 16 (group 4). Of the six items with the steepestslopes (i.e., group 1), none was classified in the last fiveranks over the 1,000 bootstraps. Although items from group2 occasionally made the top five, items from the other groupswere never or rarely classified in the top six.

-3 -2 -1 0 1 2 3

0.2

0.3

0.4

0.5

0.6

0.7

Stan

dard

err

or o

f m

easu

rem

ent

Avoidance continuum ( )θ

Figure 1. The distribution of the standard error ofmeasurement at each point along the latent avoidancecontinuum (graphed in standard deviation units), depend-ing on the number of items included in the subscale.The fifth curve from the top is for the six-item avoidancesubscale of the ECR-12 (the first curve is equivalent totwo items).

Table 3. Rank of items assessing attachment anxiety based on estimated slopes using 1,000 bootstrap samples

Item

Rank 6 8 2 14 18 24 20 4 30 26 22 36 32 10 28 12 34 16

18 0 0 0 0 0 0 0 0 0 0 0 0 0 6 22 10 110 85217 0 0 0 0 0 0 0 0 0 0 0 0 1 25 134 332 391 11716 0 0 0 0 0 0 0 0 0 0 0 0 6 96 244 344 286 2415 0 0 0 0 0 0 0 0 0 0 6 1 24 231 367 218 147 614 0 0 0 0 0 0 0 0 0 0 28 9 99 519 196 87 61 113 0 0 0 0 0 0 0 0 0 30 229 118 476 103 30 9 5 012 0 0 0 0 0 0 1 11 2 77 201 441 243 17 7 0 0 011 0 0 0 0 0 2 11 22 70 214 249 314 115 3 0 0 0 010 0 0 0 0 0 3 59 115 224 332 143 93 31 0 0 0 0 09 0 0 0 0 1 9 191 235 255 206 79 19 5 0 0 0 0 08 0 0 0 0 1 68 316 246 227 96 42 4 0 0 0 0 0 07 0 0 0 0 11 188 327 241 181 36 15 1 0 0 0 0 0 06 0 0 0 4 84 643 94 120 39 8 8 0 0 0 0 0 0 05 0 1 4 78 821 82 1 10 2 1 0 0 0 0 0 0 0 04 0 65 117 738 75 5 0 0 0 0 0 0 0 0 0 0 0 03 13 363 492 127 5 0 0 0 0 0 0 0 0 0 0 0 0 02 41 545 261 51 2 0 0 0 0 0 0 0 0 0 0 0 0 01 946 26 26 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Note. Items in bold are the anxiety items retained in the ECR-S (Wei et al., 2007).

6 M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12

European Journal of Psychological Assessment 2015 � 2015 Hogrefe Publishing

Table 4 presents the slopes’ means and confidence inter-vals for all 18 items of the anxiety subscale, in decreasingorder. The averaged item characteristic slopes were all fairlyhigh, although the six items from group 1 appeared to be thebest ones to include in a short version of the anxiety subscale.Item difficulty parameters for the 18 anxiety items, also shownin Table 4, revealed that these six items had an adequate loca-tion on the entire anxiety scale continuum (from �2.44 to2.77) and also had the lowest standard errors (< .16).

To further validate our choice of a six-item brief anxietysubscale, we compared different-length versions of the anx-iety subscale (from a 2-item to a 17-item version) based ontheir distribution of standard errors on the latent anxietycontinuum. Results indicated that the six-item version hasa standard error curve comparable to the curves for versionscomprising seven or more items, which also supported asix-item ECR anxiety subscale (Figure 2). We also com-pared the bivariate correlations between the original ECRanxiety subscale and the brief versions of different lengths.The correlation between the ECR anxiety subscale and the1-item version was .624. This correlation went up to .774with the two-item version, .813 with the three-item version,.840 with the four-item versions, .873 with the five-itemversion, and reached .880 with the six-item version, whichis adequate for an abridged version. In sum, 6 of the origi-nal 18 ECR anxiety items displayed characteristics impor-tant for inclusion in an abridged version of this subscale(items 2, 6, 8, 14, 18, and 24). Of these six items, onlytwo (items 6 and 18) were included in Wei et al.’s (2007)short version of the ECR.

Differential Item Functioning

The software program TESTGRAF was used to perform dif-ferential item functioning analyses (DIF; Ramsay, 2000) onall 2,066 participants for the 12 selected items. DIF was con-ducted to determine whether there was a gender difference inthe probability of endorsing a given item on either of the twosubscales. The characteristic curves for each item revealedthat men and women endorsed them similarly. The compos-ite DIFs (i.e., the summary statistics for the level of bias ineach item) were all less than .10 (a criterion suggested byRamsay, 2000). Thus, there is no gender bias in reportinganxiety or avoidance on the brief ECR-12. This was notthe case, however, for 5 out of 12 items included in theECR-S (Wei et al., 2007): items 11, 13, 16, 17, and 26 showedcomposite DIFs higher than .10.

Brief Discussion

The objective of Study 1 was to use two IRT-based criteriato select the best possible items for a short version ofthe ECR. Based on IRT analyses and 1,000 nonparametricbootstrap samples, six items from the original ECR avoid-ance subscale (29, 27, 25, 31, 9, and 15) and six itemsfrom the original anxiety subscale (items 6, 8, 2, 14, 18,and 24) were selected. These 12 items possessed the best

discrimination and difficulty levels, meaning that the result-ing subscales have the greatest discriminating power alonga large part of the avoidance and anxiety continuums. Thiswas rarely the case for the 12 items included in the ECR-S(Wei et al., 2007). More specifically, the IRT analysesrevealed that 9 items (11, 13, 16, 17, 22, 26, 32, 33, and35) from the 12 proposed by Wei et al. (2007) did not pos-sess optimal discriminating power.

Based on a literature review, Wei et al. (2007) suggestedthat the attachment anxiety and avoidance subscales shouldeach represent three theoretical components of avoidanceand anxiety. They used this theoretical rationale as a toolto develop the ECR-S. The items chosen here for the newECR-12 are also representative of the three componentsof anxiety – fear of interpersonal rejection (items 2, 8,and 14), disproportionate need for approval from others(18), and distress when one’s partner is unavailable (6 and24). In addition, the ECR-12 includes two of the three com-ponents of avoidance, namely excessive need for self-reliance (29 and 31) and reluctance to self-disclose (9,15, 25, and 27). No item in the ECR-12 directly tappedthe fear of interpersonal intimacy component, althoughitems 9 and 15 refer to discomfort about opening up orsharing private thoughts and feelings with others. Overall,the ECR-12 represents the various issues involved in attach-ment anxiety and avoidance, although it addresses the issueof discomfort with closeness primarily through items refer-ring to self-disclosure and openness.

Results from the DIF also showed that the new ECR-12,as opposed to the ECR-S, is not affected by a genderresponse bias (i.e., there is no difference in the way menand women endorse each item). This is important becausemany studies using the ECR involve assessing both mem-bers of heterosexual couples.

Our promising new short version needed to be evaluatedfor validity and reliability in comparison with the originalECR. In subsequent studies, different types of validityand reliability were examined, based on CTT, to further testthe psychometric properties of the new short ECR in avariety of samples.

Studies 2 Through 5: ClassicalTest Theory

The goals of Studies 2 through 5 were to examine the psy-chometric properties of the ECR-12 using CTT, to demon-strate the equivalence of the short form and the originalversion, and to demonstrate the superiority of the ECR-12over the ECR-S in a variety of samples. The specific objec-tives were (1) to confirm the factor structure of the ECR-12using CFA and test its gender invariance; (2) to assess thecorrelation between the avoidance and anxiety subscales;(3) to evaluate internal consistency and test-retest reliabilityover a 1-year period; and (4) to assess convergent and pre-dictive validity with measures of relationship satisfactionand psychological distress.

M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12 7

� 2015 Hogrefe Publishing European Journal of Psychological Assessment 2015

Table 4. Slopes’ means, standard errors, and 99% confidence intervals, and item difficulty coefficients for the anxiety subscale items

Item a (SE) 99% CI b1 (SE) b2 (SE) b3 (SE) b4 (SE) b5 (SE) b6 (SE) b7 (SE)

6. I worry that romantic partnerswon’t care about me as much as Icare about them.

5.75 (.05) [5.58–5.85] �1.27 (.07) �.89 (.05) �.34 (.03) .10 (.03) .74 (.04) 1.58 (.08) 2.04 (.10)

8. I worry a fair amount aboutlosing my partner.

5.63 (.07) [5.45–5.78] –1.60 (.08) –1.17 (.06) �.49 (.04) .04 (.03) .73 (.04) 1.62 (.08) 2.11 (.11)

2. I worry about being abandoned. 5.62 (.07) [5.40–5.78] �1.85 (.10) �1.42 (.07) �.79 (.05) �.28 (.03) .49 (.04) 1.42 (.07) 1.90 (.10)14. I worry about being alone. 5.54 (.07) [5.35–5.71] �1.21 (.07) �.84 (.05) �.26 (.03) .27 (.03) 1.01 (.06) 1.84 (.09) 2.24 (.12)18. I need a lot of reassurance that I

am loved by my partner.5.40 (.09) [5.13–5.59] �2.44 (.13) �1.97 (.10) �1.19 (.07) �.50 (.04) .33 (.03) 1.31 (.07) 1.83 (.09)

24. If I can’t get my partner toshow interest in me, I get upsetor angry.

5.24 (.11) [4.92–5.51] �1.57 (.09) �1.14 (.07) �.48 (.04) .15 (.03) 1.05 (.06) 2.17 (.11) 2.77 (.15)

20. Sometimes I feel that I force mypartners to show more feeling,more commitment.

5.11 (.11) [4.77–5.37] �1.15 (.07) �.71 (.05) �.02 (.03) .60 (.04) 1.51 (.08) 2.54 (.13) 3.04 (.17)

4. I worry a lot about myrelationships.

5.10 (.11) [4.82–5.36] �1.70 (.10) �1.09 (.06) �.24 (.04) .43 (.04) 1.32 (.07) 2.33 (.12) 4.55 (.42)

30. I get frustrated when my partner isnot around as much as I would like.

5.05 (.12) [4.72–5.36] �2.00 (.11) �1.48 (.08) �.62 (.05) .08 (.03) 1.07 (.06) 2.21 (.11) 4.51 (.42)

26. I find that my partner (s)don’t want to get as close as Iwould like.

4.96 (.13) [4.60–5.27] �.74 (.05) �.27 (.03) .44 (.04) 1.03 (.06) 1.83 (.10) 2.78 (.15) 3.29 (.19)

22. I do not often worry about beingabandoned.

4.83 (.16) [4.36–5.19] �2.23 (.13) �1.57 (.09) �.56 (.05) .17 (.04) .99 (.04) 2.18 (.12) 2.93 (.16)

36. I resent it when my partnerspends time away from me.

4.80 (.13) [4.45–5.09] �1.75 (.10) �1.21 (.07) �.30 (.04) .55 (.04) 1.61 (.09) 2.68 (.14) 3.18 (.18)

32. I get frustrated if romanticpartners are not available when Ineed them.

4.69 (.16) [4.29–5.08] �2.33 (.13) �1.79 (.10) �.86 (.06) �.03 (.03) 1.03 (.06) 2.29 (.12) 2.90 (.16)

10. I often wish that my partner’sfeelings for me were as strong asmy feelings for him/her.

4.42 (.15) [3.96–4.77] �3.26 (.20) �2.81 (.17) �2.11 (.13) �1.35 (.10) �.40 (.05) .57 (.05) 1.12 (.08)

28. When I’m not involved in arelationship, I feel somewhatanxious and insecure.

4.29 (.15) [3.85–4.63] �1.15 (.08) �.64 (.06) .13 (.05) .93 (.07) 2.05 (.12) 3.28 (.19) 3.94 (.24)

12. I often want to merge completelywith romantic partners, and thissometimes cares them away.

4.21 (.17) [3.71–4.63] �.42 (.05) .12 (.04) .92 (.06) 1.66 (.10) 2.60 (.15) 3.55 (.21) 4.04 (.26)

34. When romantic partnersdisapprove of me, I feel reallybad about myself.

4.16 (.17) [3.66–4.59] �2.37 (.15) �1.68 (.11) �.55 (.06) .48 (.05) 1.77 (.10) 3.10 (.18) 3.74 (.22)

16. My desire to be very closesometimes scares people away.

3.90 (.18) [3.42–4.31] �.33 (.05) .27 (.04) 1.14 (.07) 1.87 (.11) 2.88 (.17) 3.94 (.24) 4.42 (.30)

Notes. Items in bold are the anxiety items retained for the ECR-12. Italicized items are the anxiety items retained in the ECR-S (Wei et al., 2007).

8M

.-F.L

afontaineet

al.:P

sychometrics

Properties

ofthe

EC

R-12

European

Journalof

Psychological

Assessm

ent2015

�2015

Hogrefe

Publishing

Method

Study 2

The second sample was composed of 316 French-Canadianheterosexual couples from the province of Quebec (marriedn = 202, cohabiting n = 114). A survey firm used random-digit dialing to recruit adult participants who had been mar-ried or cohabiting for at least 6 months. The mean age ofthe participants was 39 years, with an average of 14 yearsof education, and average relationship duration of 13 years.Most couples had children (73.7%). Questionnaire packetswith prepaid return envelopes were mailed to 500 couples(two separate packets to ensure confidentiality). Of the500 couples, 316 completed and returned both partners’questionnaires (response rate = 63.2%). A year later, 154couples agreed to complete the questionnaires again(response rate = 48.7%). All participants completed theFrench-language version of the original ECR at Time 1and Time 2. They also completed the four-item French ver-sion of the Dyadic Adjustment Scale (DAS-4; Spanier,1976; abridged by Sabourin, Valois, & Lussier, 2005)assessing couple satisfaction, and the French-language briefPsychological Symptoms Index (PSI; Ilfeld, 1976) assess-ing psychological distress. The DAS-4 Cronbach’s alphaswere .83 for women and .79 for men, whereas the PSI alphacoefficients were .90 for women and .89 for men.

Study 3

The third sample included 224 English-speaking Canadianheterosexual couples (28.8% married) recruited from the

general community. The mean age of the participants was30 years for women and 33 years for men. The majorityof women (82.8%) and men (78.4%) were college educated.Couples had been living together for an average of 6 years,and 30.2% had children. Couples were recruited throughlocal newspapers, posters, and community events in theprovince of Ontario. At Time 1, they took part in a largerstudy requiring them to participate in a 2-hr testing session.Partners completed a questionnaire package independentlyfrom one another. One year later, 116 couples (responserate = 51.8%) agreed to complete the questionnaires a sec-ond time. At Time 1, participants completed the originalversion of the ECR, as well as the English version of theDAS-4 (women, a = .76; men, a = .73). At Time 2, partic-ipants completed the original version of the ECR, as wellas the DAS-4 (women, a = .76; men, a = .87) and thePSI (women, a = .91; men, a = .92).

Study 4

The fourth sample included 107 men and 288 womeninvolved in same-sex relationships, recruited in the provinceof Ontario, Canada. They identified themselves as gay orlesbian (78%) or as bisexual (22%). To participate, individ-uals had to be at least 18 years old, and to have been in theircurrent same-sex couple relationship for at least 12 months.Mean age of the participants was 29 years and mean lengthof their current relationship was 4.18 years. Most of the par-ticipants had a college degree (72.9%) and 12.9% had chil-dren. They were recruited through flyers, emails, or postersin local organizations serving the gay and lesbian (GLBT)community. Persons interested in participating in the studywere sent a link to the online survey or a paper question-naire. They completed the original ECR and the Englishversion of the DAS-4 (a = .81).

Study 5

The fifth sample included 524 Canadian couples (40.6%married) seeking couple therapy in a private practice ineither Quebec or Ontario. The couples spoke French(83.4%) or English (16.6%). The mean age was 40 yearsfor women and 43 years for men. The majority of thewomen (84.0%) and the men (88.2%) were collegeeducated. Couples had been living together for an averageof 12 years, and 83.2% had children. Partners were invitedto independently complete the ECR, DAS-4(women, a = .74; men, a = .74), and the PSI measure ofpsychological distress (women, a = .91; men, a = .91) justbefore the first therapy session.

Results

Factor Structure

Confirmatory Factor Analyses (CFA) were conducted sepa-rately for women and men to confirm the factor structure of

-3 -2 -1 0 1 2 3

0.3

0.5

0.4

0.6

0.7

Stan

dard

err

or o

f m

easu

rem

ent

Anxiety continuum ( )θ

Figure 2. The distribution of the standard error ofmeasurement at each point along the latent anxietycontinuum (graphed in standard deviation units), depend-ing on the number of items included in the subscale. Thefifth curve from the top is for the six-item anxietysubscale of the ECR-12 (the first curve is equivalent totwo items).

M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12 9

� 2015 Hogrefe Publishing European Journal of Psychological Assessment 2015

the ECR-12 and ECR-S, using AMOS (Arbuckle, 1999).Maximum likelihood estimation was employed to estimatethe two-factor models, with items 29, 27, 25, 31, 9, and 15predicted by the ECR-12 avoidance latent factor(ECR-S = items 11, 13, 17, 27, 33, 35) and items 6, 8, 2,14, 18, and 24 predicted by the ECR-12 anxiety latent fac-tor (ECR-S = items 6, 16, 18, 22, 26, 32). When Brennanet al. (1998) created the ECR, the two 18-item scales werehypothesized to be uncorrelated. However, Cameron,Finnegan, and Morry (2012) have recommended reportingthe correlation between anxiety and avoidance based on ameta-analysis. In the present study we estimated the covari-ance between the two factors. A covariance was also addedbetween the error terms of items 29 and 31 to account forsimilar wording.

Each CFA was tested using several fit indices, namelythe Comparative Fit Index (CFI), the Tucker Lewis Index(TLI), and the Root Mean Square Error of Approximation(RMSEA). The CFI and TLI values typically range from0 to 1, and values equal to or greater than .90 are represen-tative of a well-fitting model. The RMSEA calculates theaverage error of fit between the submitted model and thedata, and a value equal to or lower than .08 correspondsto an adequate fit (Browne & Cudeck, 1993). The chi-square test is perceived to be less valid with a larger samplesize. It is therefore preferable to use the v2/df ratio. Whenthe value of the ratio falls between 1 and 5, the fit betweenthe proposed theoretical model and the data is satisfactory(Bollen, 1989). As shown in Table 5, the indexes of fit weremarginal or adequate for both women and men with regardto the ECR-12. In contrast, the CFA yielded poorer andinadequate indexes of fit for the ECR-S in all samples, sug-gesting poor factorial structure for the ECR-S.

Across all samples, the correlation between the anxietyand avoidance latent factors was closer to the expected nullcorrelation for the ECR-12 than it was for the ECR-S. InStudy 2, this correlation was small for the ECR-12 (women:r = .07, p = .346; men: r = .16, p = .034), whereas for theECR-S it was moderate for women (r = .43, p < .001) andsurprisingly strong for men (r = .71, p < .001). In Study 3,the correlation between the ECR-12 anxiety and avoidancefactors was .17 for women ( p = .059) and .14 for men( p = .132), whereas for the ECR-S it reached .29 forwomen ( p = .001) and .35 for men ( p < .001). In Study4, this correlation was small for the ECR-12 (r = .13,p = .050) and moderate for the ECR-S (r = .38,p < .001). In Study 5, the correlation between the ECR-12 anxiety and avoidance factors was close to zero forwomen (r = .01, p = .841) and men (r = .04, p = .485),whereas for the ECR-S it was significant for men(r = .29, p < .001) but not for women (r = .10, p = .094).

Structure Invariance

Comparisons of the equivalence of factor loadings betweenmen and women were conducted, using the multiple groupcomparison approach (Byrne, 2001).1 In Study 2, signifi-cant chi-square difference tests indicated a differencebetween men and women in factor loading estimates forthe ECR-12 (v2(10, N = 632) = 21.98, p = .015) and theECR-S (v2(10, N = 632) = 39.49, p < .001). Genderinvariance testing of the ECR-12 revealed no significantdifferences between men and women in the factor loadingsin Study 3 (v2(10, N = 568) = 7.46, p = .682), Study 4(v2(10, N = 392) = 13.60, p = .192), and Study 5 (v2(10,

Table 5. Fit indexes for the Confirmatory Factor Analyses of the ECR-12 and the ECR-S

Study Version v2 v2/df CFI TLI RMSEA

2 W ECR-12 v2(52, N = 316) = 130.81, p < .001 2.52 .92 .88 .07ECR-S v2(53, N = 316) = 192.40, p < .001 3.63 .77 .66 .09

2 M ECR-12 v2(52, N = 316) = 136.97, p < .001 2.63 .92 .87 .07ECR-S v2(53, N = 316) = 226.53, p < .001 4.27 .75 .64 .10

3 W ECR-12 v2(52, N = 284) = 116.37, p < .001 2.23 .93 .89 .07ECR-S v2(53, N = 284) = 209.89, p < .001 3.96 .83 .75 .10

3 M ECR-12 v2(52, N = 284) = 119.74, p < .001 2.30 .90 .84 .07ECR-S v2(53, N = 284) = 152.98, p < .001 2.89 .84 .76 .08

4 M&W ECR-12 v2(52, N = 392) = 132.08, p < .001 2.54 .95 .92 .06ECR-S v2(53, N = 392) = 321.69, p < .001 6.07 .82 .74 .11

5 W ECR-12 v2(52, N = 524) = 109.66, p < .001 2.11 .97 .96 .05ECR-S v2(53, N = 524) = 300.95, p < .001 5.68 .79 .70 .09

5 M ECR-12 v2(52, N = 524) = 172.58, p < .001 3.32 .94 .90 .07ECR-S v2(53, N = 524) = 295.27, p < .001 5.57 .77 .66 .09

Note. W = Women; M = Men.

1 We also tested language invariance of the ECR-12 and ECR-S separately for men and women (because of the aforementionednonindependence of the dyadic data) using the two community samples of couples (Studies 2 and 3). Language invariance testing of theECR-12 revealed no significant differences in the factor loadings between the French- and English-speaking women (v2(10,N = 600) = 11.368, p = .330) and men (v2(10, N = 600) = 8.699, p = .561). For the ECR-S, a nonsignificant chi-square difference testwas found in men (v2(10, N = 600) = 14.711, p = .143), but a significant language difference was found in women (v2(10,N = 600) = 32.556, p < .001).

10 M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12

European Journal of Psychological Assessment 2015 � 2015 Hogrefe Publishing

N = 1054) = 11.64, p = .310). For the ECR-S, a nonsignif-icant chi-square difference test was found in Study 3 (v2(10,N = 568) = 15.23, p = .124), but significant gender differ-ences were found in Study 4 (v2(10, N = 392) = 22.52,p = .013) and Study 5 (v2(10, N = 1,058) = 22.28,p = .014).

Correlations With Original ECR

The correlations between the brief and the original sub-scales of the ECR were also tested to ensure representative-ness of the new short scale. Across all samples (Study 2–5),the original anxiety subscale correlated highly with theshort anxiety subscale from both the ECR-12 (rs from .89to .95) and ECR-S (rs from .86 to .94). Similar results werealso found for the short avoidance subscales from both theECR-12 (rs from .84 to .92) and ECR-S (rs from .87 to.93). Z tests revealed no significant differences betweenthe correlations obtained for the ECR-S and the ECR-12(all ps > .05), suggesting equivalence of the short versions.

Test-Retest Stability

Test-retest stability of the ECR, ECR-12, and ECR-Sover a 1-year period was examined in two samples

(Studies 2 and 3). Using AMOS, structural equation model-ing (SEM) allowed us to estimate regression coefficientsbetween Time 1 and Time 2 for the original and the twobrief ECR versions. To create indicators of the latent anxi-ety and avoidance variables, the six items on each ECR sub-scale were randomly divided into three parcels andaveraged (Little, Cunningham, Shahar, & Widaman,2002). According to Sibley et al. (2005), the use of SEM toassess stability reduces the effect of measurement error andprovides a better appraisal of the underlying construct. To ruleout the possibility of gender differences and to acknowledgethe nonindependence of the dyadic data, separate analyseswere conducted for men and women. In both samples,gender-specific results revealed that Time 1 latent avoidancepredicted Time 2 latent avoidance, and Time 1 latent anxietypredicted Time 2 latent anxiety (see Table 6). Similar esti-mates were found for the abridged and original subscales ofthe ECR, suggesting equivalence of the short versions.

Internal Consistency

Cronbach’s alpha coefficients in the subscales of the originaland brief versions of the ECR are displayed in Table 7 for allsamples. Cronbach’s alphas for the ECR subscales and theECR-12 subscales were all adequate, whereas the coeffi-cients were poorer for the ECR-S (a < .70) in Studies 2and 5 (especially for the men’s avoidance subscale).

Convergent and Predictive Validity

To examine the convergent validity of the brief version ofthe ECR, regression analyses (considering both anxietyand avoidance in the same analysis, as recommended byCameron et al., 2012) predicting relationship satisfactionand psychological distress were performed. The resultsrevealed significant associations between the attachmentsubscales (both the original and the brief versions) andmeasures of psychological distress and relationship satisfac-tion (see Table 8). In the clinical sample of Study 5, how-ever, only the avoidance subscales (negatively) predictedrelationship satisfaction. Z tests revealed no significant dif-ferences between the coefficients obtained with the originalECR, the ECR-S, and the ECR-12 (all ps > .05), suggestingequivalence of the short versions.

Table 6. Standardized coefficients (bs) from test-reteststructural equation modeling

Women Men

Study 2ECR-12 anxiety .82*** .80***ECR-S anxiety .96*** .89***ECR anxiety .86*** .80***ECR-12 avoidance .67*** .70***ECR-S avoidance .62*** .78***ECR avoidance .67*** .70***

Study 3ECR-12 anxiety .67*** .81***ECR-S anxiety .65*** .89***ECR anxiety .69*** .81***ECR-12 avoidance .53*** .65***ECR-S avoidance .67*** .66***ECR avoidance .64*** .64***

Note. ***p < .001.

Table 7. Cronbach’s alpha coefficients and 95% confidence intervals for ECR anxiety and avoidance subscales

Study NECR-12anxiety

ECR-Sanxiety

ECRanxiety

ECR-12avoidance

ECR-Savoidance

ECRavoidance

2 W 316 .78 [.74; .82] .60 [.53; .67] .86 [.84; .88] .75 [.71; .79] .73 [.68; .78] .88 [.86; .90]2 M 316 .78 [.74; .82] .64 [.57; .70] .86 [.83; .88] .80 [.77; .84] .69 [.63; .74] .87 [.85; .89]3 W 224 .84 [.81; .88] .77 [.71; .81] .91 [.89; .92] .83 [.79; .86] .86 [.83; .89] .93 [.92; .94]3 M 224 .82 [.78; .86] .75 [.69; .80] .90 [.88; .92] .74 [.67; .79] .79 [.74; .83] .89 [.87; .91]4 M&W 392 .87 [.84; .89] .82 [.79; .85] .93 [.92; .94] .79 [.75; .82] .82 [.79; .85] .90 [.88; .92]5 W 524 .83 [.80; .85] .71 [.67; .74] .89 [.87; .90] .82 [.79; .84] .71 [.67; .75] .89 [.88; .91]5 M 524 .80 [.78; .83] .73 [.69; .77] .88 [.86; .89] .80 [.78; .83] .64 [.59; .69] .88 [.86; .89]

Note. W = Women; M = Men.

M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12 11

� 2015 Hogrefe Publishing European Journal of Psychological Assessment 2015

Discussion

The main objective of this research project was to developan optimal brief version of the frequently used but ratherlong Experiences in Close Relationships (ECR) measure(Brennan et al., 1998), using IRT analyses to evaluate itemperformance. The resulting self-report questionnaire iscalled the ECR-12. A secondary objective was to determinethe psychometric properties of the ECR-12 in couples fromthe general community (Studies 2 and 3), individualsinvolved in same-sex relationships (Study 4), and dyadsseeking couple therapy (Study 5). The five studies showthat the ECR-12 preserves the good psychometric proper-ties of the ECR-36 and generally has better psychometricproperties than the previously developed brief measure,the ECR-S (Wei et al., 2007).

Development of the ECR-12

Using two criteria from IRT, six items from the originalECR avoidance subscale (9, 15, 25, 27, 29, and 31) andsix items from the original anxiety subscale (items 2, 6,8, 14, 18, and 24) were selected, preserving the theoreticallygrounded two-dimensional structure of the original ECR(Brennan et al., 1998). These 12 items possessed the bestitem discrimination and difficulty indices, meaning thatthe resulting subscales had the greatest discriminatingpower along a large portion of the avoidance continuumand all along the anxiety continuum. This was not always

the case for the 12 items included in the ECR-S, which alsoevidenced higher measurement errors.

Psychometric Properties of the ECR-12

The two-dimensional structure of the ECR-12 was wellestablished with confirmatory factor analyses (CFA) in allfive samples. The distinction between attachment anxietyand avoidance was clear across a variety of populations,subcultures, genders, and languages. We used severalindexes to assess the model’s goodness of fit, one of whichwas the comparative fit index (CFI). CFI values greaterthan .90 indicate a reasonable fit of the data, and valuesof .95 or greater indicate a model that fits the data well(Hu & Bentler, 1999). The ECR-12 consistently fit the pre-dicted two-dimensional data structure much better than Weiet al.’s (2007) ECR-S in all four Canadian samples (French-speaking couples, English-speaking couples, individualsin same-sex relationships, and couples seeking therapy).In comparison to the ECR-12, poor and inadequate indexesof fit were obtained for the ECR-S in all samples, suggest-ing poorer factorial structure for the ECR-S. Wei et al.(2007) had to create two additional latent variables(accounting for response set) to improve the goodness offit of the model to their data. The ECR-12, like the originalECR, did not require the creation of such latent variables,even with its use of reverse-scored items.

The statistical independence of the ECR-12’s two sub-scales (avoidance and anxiety) was generally demonstratedacross our diverse samples. In cases where avoidance and

Table 8. Standardized regression coefficients from analyses in which ECR subscales (attachment anxiety and avoidance)predicted two validity criteria, relationship satisfaction (DAS-4) and psychological distress (PSI)

Validitycriteria

ECR-12anxiety

ECR-Sanxiety

ECRanxiety

ECR-12avoidance

ECR-Savoidance

ECRavoidance

Study 2DAS-4 W �.21*** �.23*** �.20*** �.49*** �.50*** �.54***DAS-4 M �.09 �.04 �.04 �.47*** �.53*** �.56***PSI W .23*** .25*** .26*** .34*** .30*** .33***PSI M .28*** .30*** .33*** .21*** .19** .21***

Study 3DAS-4 W �.13 �.09 �.13 �.56*** �.63*** �.52***DAS-4 M �.10 �.06 �.09 �.52*** �.46*** �.39**DAS-4 T2W �.08 �.06 �.06 �.50*** �.54*** �.51***DAS-4 T2M �.25 �.26 �.19 �.23 �.30* �.33*PSI T2W .38** .32* .33* .08 .26 .26PSI T2M .48*** .44** .43** .07 .12 .14

Study 4DAS-4 �.22** �.20** �.18** �.34*** �.42*** �.41***

Study 5DAS-4 W .09 .04 .06 �.39** �.36*** �.39***DAS-4 M �.05 �.07 �.08* �.42*** �.43*** �.45***PSI W .30*** .27*** .31*** .09* .11* .10*PSI M .32*** .30*** .34*** .20*** .19*** .23***

Notes. *p < .05. **p < .01. ***p < .001. W = Women; M = Men; DAS-4 = Dyadic Adjustment Scale (four-item short form);PSI = Psychological Symptoms Index; T2 = Time 2.

12 M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12

European Journal of Psychological Assessment 2015 � 2015 Hogrefe Publishing

anxiety scores were significantly related, the correlationwas small, as intended by the creators of the originalECR (Brennan et al., 1998). In this regard, the ECR-S(Wei et al., 2007) is akin to the ECR-R (Fraley et al.,2000), which the meta-analysis conducted by Cameronet al. (2012) found to generate higher correlations betweenanxiety and avoidance than the original ECR. Moreover, thecorrelations found between the ECR-S anxiety and avoid-ance subscales within our samples sometimes reached.70, a threshold where a multicollinearity problem can arise.Our results suggest that this problem will not be encoun-tered with the ECR-12.

Based on the internal consistency of the questionnaireitems and the questionnaire’s stability over a 1-year period(in Studies 2 and 3), the ECR-12 proved to be a reliablemeasure in our four samples (French-speaking couples,English-speaking couples, individuals in same-sex relation-ships, and couples seeking therapy). Specifically,Cronbach’s alphas varied from .78 to .87 for the anxietysubscale and from .74 to .83 for the avoidance subscale.Due to the decrease in the number of items and the elimi-nation of redundant items, it was anticipated that the alphasfor the ECR-12 would be slightly lower than the ones forthe original ECR subscales. However, alpha coefficientsremained .74 or higher for the ECR-12, indicating accept-able to excellent internal consistency for short subscales.In comparison, Cronbach’s alphas for the ECR-S were gen-erally lower and in a few cases fell within the .60 range –among French-speaking couples and couples seeking ther-apy – suggesting higher measurement errors with thosepopulations. Findings from Studies 2 and 3 demonstratedthe relatively good stability of all three versions of theECR over a period as long as 1 year.

As expected, both the anxiety and avoidance subscalescorrelated with two validity criteria: relationship satisfac-tion and psychological distress. Specifically, findings fromStudies 2 through 5 provided further support for the conver-gent and predictive validity of the ECR-12, with correla-tions similar to those of the original ECR. These resultsare consistent with attachment theory and the results ofseveral previous studies (see Mikulincer & Shaver, 2007,for a review).

Gender Considerations

As expected, no gender differences were found in the prob-ability of endorsing each level of anxiety or avoidance oneach item on the ECR-12 (Study 1). Gender differenceswere found, however, for 5 out of 12 items from theECR-S. We also compared the factor structure and factorloadings for men and women in Studies 2 to 5. A gender-invariant factor structure of the ECR-12 was found in threeout of four samples (Studies 3, 4, and 5, but not Study 2).In couples from the community, seeking therapy, and same-sex couples the link between the items and their target fac-tor was comparable for men and women, but there was a

significant gender difference in the sample of French-Canadian couples. We found, however, gender differencesin the factor loadings for Wei et al.’s ECR-S in three outof four samples (Studies 2, 4, and 5). Language invariancetesting of the ECR-12, using the two community samples ofcouples (Studies 2 and 3), revealed no significant differ-ences in the factor loadings between the French- andEnglish-speaking women and men. However, we foundlanguage differences in the factor loadings for Wei et al.’sECR-S in the women sample. Thus, we conclude that thenew ECR-12 factor structure is more likely to be invariantwith regard to gender and language across samples than theECR-S.

Implications

Using item response theory, we selected 12 items from theoriginal ECR that exhibit good measurement precision atmost levels of the anxiety and avoidance continuums. Fourstudies demonstrated the reliability and convergent andpredictive validities of the ECR-12. In practical terms,we reduced the number of items from 36 to 12 withoutcompromising the desirable psychometric properties ofthe original ECR. Moreover, the high correlations betweenthe original ECR subscales and the brief versions of them(rs around .90) imply that the two versions of the sub-scales assess the same constructs. The ECR-12 maintainsthe brevity of the ECR-S developed by Wei et al. (2007),but was developed using an IRT method similar to that em-ployed in the development of the ECR-R (Fraley et al.,2000). In the present studies, the good psychometric prop-erties of the ECR-12 were demonstrated in samples ofshort-term and long-term relationships, heterosexual andsame-sex relationships, and couples from different settings(college, community, clinics) and with participantswho spoke different languages (French and English).The ECR-12 can confidently be used by researchers andmental health practitioners among these populations whena short measure of attachment anxiety and avoidance isneeded.

Four limitations of our studies are that (a) all of the datacame from self-report measures, (b) the only validity crite-ria were relationship satisfaction and psychological distress,(c) each six-item scale contains a mix of response to roman-tic partners in general and relationship-specific items as inthe original ECR, and (d) there was an overlap in the word-ing for two of the items included in the ECR-12 due to thefact that item selection was mainly based on the IRT anal-yses. In future studies it will be important to use otherassessment methods such as behavioral observations,social-cognitive tasks, daily diaries, and physiological mea-surements to further establish the validity of the ECR-12.Because all of these methods have already been used withthe original ECR (see review by Mikulincer & Shaver,2007), and given that the ECR-12 is structured and func-tions much like the original ECR, we are confident that

M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12 13

� 2015 Hogrefe Publishing European Journal of Psychological Assessment 2015

the ECR-12 will prove to have high construct validity in fu-ture studies.

References

Ainsworth, M. D. S., Blehar, M. M., Waters, E., & Wall, S.(1978). Patterns of attachment: A psychological study of thestrange situation. Hillsdale, NJ: Erlbaum.

Alonso-Arbiol, I., Balluerka, N., & Shaver, P. R. (2007). ASpanish version of the Experiences in Close Relationships(ECR) adult attachment questionnaire. Personal Relation-ships, 14, 45–63.

Arbuckle, J. (1999). AMOS 4.0 [Computer software]. Chicago,IL: Smallwaters.

Baker, F. B. (2001). The basics of item response theory. CollegePark, MD: ERIC Clearinghouse on Assessment andEvaluation.

Bartholomew, K., & Horowitz, L. M. (1991). Attachment stylesamong young adults: A test of a four-category model.Journal of Personality and Social Psychology, 61, 226–244.

Bollen, K. A. (1989). Structural equations with latent variables.New York, NY: Wiley.

Brennan, K. A., Clark, C. L., & Shaver, P. R. (1998). Self-reportmeasurement of adult attachment: An integrative overview.In J. A. Simpson & W. S. Rholes (Eds.), Attachment theoryand close relationships (pp. 46–76). New York, NY: GuilfordPress.

Browne, M. W., & Cudeck, R. (1993). Alternative ways ofassessing model fit. In K. A. Bollen & J. S. Long (Eds.),Testing structural equation models (pp. 136–162). BeverlyHills, CA: Sage.

Bowlby, J. (1969/1982). Attachment and loss: Vol. 1. Attach-ment. New York, NY: Basic Books.

Byrne, B. M. (2001). Structural equation modeling with Amos:Basic concepts, applications, and programming. Mahwah,NJ: Erlbaum.

Cameron, J., Finnegan, H., & Morry, M. (2012). Orthogonaldreams in an oblique world: A meta-analysis of theassociation between attachment anxiety and avoidance.Journal of Research in Personality, 46, 472–476.

Collins, N. L., & Allard, L. M. (2001). Cognitive representa-tions of attachment: The content and function of workingmodels. In G. J. O. Fletcher & M. S. Clark (Eds.), Blackwellhandbook of social psychology: Vol. 2. Interpersonalprocesses (pp. 60–85). Oxford, UK: Blackwell.

Conradi, H. J., Gerlsma, C., van Duijn, M., & de Jonge, P.(2006). Internal and external validity of the Experiences inClose Relationships Questionnaire in an American and twoDutch samples. European Journal of Psychiatry, 20,258–269.

Del Giudice, M. (2011). Sex Differences in romantic attach-ment: A meta-analysis. Personality Social PsychologyBulletin, 37, 193–214.

Embretson, S. E., & Reise, S. P. (2000). Item response theory forpsychologists. Mahwah, NJ: Erlbaum.

Fairchild, A., & Finney, S. J. (2006). Investigating validityevidence for the Experiences in Close Relationships-Revisedquestionnaire. Educational and Psychological Measurement,66, 116–135.

Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000). An itemresponse theory analysis of self-report measures of adultattachment. Journal of Personality and Social Psychology,78, 350–365.

Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexesin covariance structure analysis: Conventional criteria versusnew alternatives. Structural Equation Modeling, 6, 1–55.

Ilfeld, F. W. (1976). Further validation of a psychiatric symptomin a normal population. Psychological Reports, 39, 1215–1228.

Lafontaine, M.-F., & Lussier, Y. (2003). Bidimensional structureof attachment in love: Anxiety over abandonment andavoidance of intimacy. Canadian Journal of BehaviouralScience, 35, 56–60.

Little, T. D., Cunningham, W. A., Shahar, G., & Widaman, K. F.(2002). To parcel or not to parcel: Exploring the question,weighing the merits. Structural Equation Modeling, 9,151–173.

MacDonald, P. L., & Paunonen, S. V. (2002). A Monte Carlosimulation of item and person statistics based on itemresponse theory and classical test theory. Educational andPsychological Measurement, 62, 921–943.

Mallinckrodt, B., & Wang, C. (2004). Quantitative methods forverifying semantic equivalence of translated research instru-ments: A Chinese version of the Experiences in CloseRelationships Scale. Journal of Counseling Psychology, 51,368–379.

Marsh, H. W., Hau, K.-T., Balla, J. R., & Grayson, D. (1998). Ismore ever too much? The number of indicators per factor inconfirmatory factor analysis. Multivariate BehavioralResearch, 33, 181–230.

Matte, M., Lemieux, A., & Lafontaine, M.-F. (2009). Briefreport: Factor structure of the Experiences in Close Rela-tionships with gay and lesbian individuals. North AmericanJournal of Psychology, 11, 361–368.

Mikulincer, M., & Florian, V. (2000). Exploring individualdifferences in reactions to mortality salience: Does attach-ment style regulate terror management mechanisms? Jour-nal of Personality and Social Psychology, 79, 260–273.

Mikulincer, M., & Shaver, P. R. (2007). Attachment in adult-hood: Structure, dynamics, and change. New York, NY:Guilford Press.

Mooney, C., & Duval, R. (1993). Bootstrapping: A nonpara-metric approach to statistical inference. Thousand Oaks,CA: Sage.

Nakao, T., & Kato, K. (2004). Constructing the Japanese versionof the Adult Attachment Style Scale (ECR). JapaneseJournal of Psychology, 75, 154–159.

Nunnally, J. C., & Bernstein, I. H. (1994). Psychometric theory.New York, NY: McGraw-Hill.

Picardi, A., Bitetti, D., Puddu, P., & Pasquini, P. (2000).Development and validation of an Italian version of thequestionnaire ‘‘Experiences in Close Relationships,’’ a newself-report measure of adult attachment. Rivista diPsichiatria, 35, 114–120.

Poitras, S.-C., Guay, F., & Ratelle, C. F. (2012). Using the self-directed search in research: Selecting a representative poolof items to measure vocational interests. Journal of CareerDevelopment, 39, 186–207.

Ramsay, J. O. (2000). TestGraf: A program for the graphical analysisof multiple choice test and questionnaire data. Unpublishedmanuscript, McGill University, Montreal, Canada.

Reeve, B. B., & Fayers, P. (2005). Applying item responsetheory modelling for evaluating questionnaire item and scaleproperties. In P. Fayers & R. D. Hays (Eds.), Assessingquality of life in clinical trials: Methods and practice (2nded., pp. 55–73). New York, NY: Oxford University Press.

Robins, R. W., Hendin, H. M., & Trzesniewski, K. H. (2001).Measuring global self-esteem: Construct validation of asingle-item measure and the Rosenberg Self-Esteem Scale.Personality and Social Psychology Bulletin, 27, 151–161.

Sabourin, S., Valois, P., & Lussier, Y. (2005). Development andvalidation of a brief version of the Dyadic Adjustment Scalewith a nonparametric item analysis model. PsychologicalAssessment, 17, 15–27.

14 M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12

European Journal of Psychological Assessment 2015 � 2015 Hogrefe Publishing

Samejima, F. (1969). Estimation of ability using a responsepattern of graded scores. Psychometrika Monograph Sup-plement, 34, 1–97.

Samejima, F. (1997). Graded response model. In W. J. van derLinden & R. K. Hambleton (Eds.), Handbook of modernitem response theory (pp. 85–100). New York, NY: Springer-Verlag.

Sibley, C. G., Fischer, R., & Liu, J. H. (2005). Reliability andvalidity of the revised Experiences in Close Relationships(ECR-R) self-report measure of adult romantic attachment.Personality and Social Psychology Bulletin, 31, 1524–1536.

Spanier, G. B. (1976). Measuring dyadic adjustment: Newscales for assessing the quality of marriage and similardyads. Journal of Marriage and the Family, 38, 15–28.

Stroud, M. W., McKnight, P. E., & Jensen, M. P. (2004).Assessment of self-reported physical activity in patientswith chronic pain: Development of an abbreviated Roland-Morris disability scale. The Journal of Pain, 5, 257–263.

Thissen, D., & Steinberg, L. (1997). A response model formultiple choice items. In W. J. van der Linden & R. K.Hambleton (Eds.), Handbook of modern item responsetheory (pp. 381–394). New York, NY: Springer.

Valois, P., Houssemand, C., Germain, S., & Abdous, B. (2011).An open source tool to verify the psychometric properties ofan evaluation instrument. Procedia, Social and BehavioralSciences, 15, 552–556.

Wei, M., Russell, D. W., Mallinckrodt, B., & Vogel, D. L.(2007). The Experiences in Close Relationships (ECR)Scale-short form: Reliability, validity, and factor structure.Journal of Personality Assessment, 88, 187–204.

Date of acceptance: August 29, 2014Published online: February 27, 2015

Marie-France Lafontaine

Department of PsychologyUniversity of Ottawa136 Jean-Jacques LussierOttawa K1N 6N5CanadaE-mail [email protected]

M.-F. Lafontaine et al.: Psychometrics Properties of the ECR-12 15

� 2015 Hogrefe Publishing European Journal of Psychological Assessment 2015