Information on the Experiences in Close Relationships-Revised (ECR-R) Adult Attachment Questionnaire |
R. Chris Fraley
University of Illinois at Urbana-Champaign
The Experiences in Close Relationships-Revised (ECR-R) questionnaire is a revised version of Brennan, Clark, and Shaver's (1998) Experiences in Close Relationships (ECR) questionnaire. The items on the ECR-R were selected using techniques based on Item Response Theory, but were selected from the same item pool as those from the ECR. Both the ECR and the ECR-R are designed to assess individual differences with respect to attachment-related anxiety (i.e., the extent to which people are insecure vs. secure about the availability and responsiveness of romantic partners) and attachment-related avoidance (i.e., the extent to which people are uncomfortable being close to others vs. secure depending on others). This web page is designed to answer some frequently asked questions about the ECR-R. More detailed information about the ECR-R can be found in the original article in which the questionnaire was published:
Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000). An item-response theory analysis of self-report measures of adult attachment. Journal of Personality and Social Psychology, 78, 350-365.If you would like a copy of this article, please contact me. If you would like to download a PDF version of the article, you can do so on the "publications" page of my web site.
For more general information on the measurement of adult attachment via self-report, please see the on-line document, Self-Report Measures of Adult Attachment, (and the references therein) by Phil Shaver and Chris Fraley.
For more information on the two-dimensional model of individual differences in adult attachment, please see the following overview of theory and research on adult attachment: A Brief Overview of Adult Attachment Theory and Research by Chris Fraley.
Frequently asked questions about the Experiences in Close Relationships-Revised measure
Q: Where can I find the ECR-R items?
A. You can find the ECR-R items in the Fraley, Waller, and Brennan (2000) JPSP paper. The items can also be easily copied-and-pasted into MS Word via the page at this link. They are also listed below.
Q: Is the ECR-R different from the on-line Close Relationships Questionnaire / Attachment Style Questionnaire that I saw on the Internet?
A. The ECR-R is the same instrument as the 'Close Relationships Questionnaire' or the 'Attachment Style Questionnaire' at http://www.web-research-design.net/cgi-bin/crq/crq.pl.
Update Dec 9, 2005: The "Attachment Style Questionnaire" that is on-line is now a modified version of the original ECR-R. Approximately 16 of the items are from the original instrument; the remainder are selected at random from an item bank of over 300 items in an attempt to identify items that can be used to improve the scales. Thus, please do not simply copy and paste the items from the online questionnaire under the assumption that those are the complete set of ECR-R items. Moreover, please do not use the online questionnaire as a method for scoring your own paper-and-pencil tests.
Q: How do I score the ECR-R?
A. For each person, average the scores for all items within each scale. The items, separated by scale, are listed below; they are also listed here. Note: It is important to "reverse key" the items denoted with an * before averaging item responses. Items such as "I rarely worry about my partner leaving me" should be keyed as counting against scores on anxiety, not for scores on anxiety. The easiest way to do this is to subtract 8 from each reverse keyed item (thus, a high score of 6 becomes a low score of 2, for example) before averaging them. You can find more information on scoring at this link
As a simple example, assume you administer the ECR-R to someone named Chris. Chris provides the following ratings to the items listed here : ECRR01=1, ECRR02=2, ECRR03=2, ECRR04=1, ECRR05=3, ECRR06=2, ECRR07=2, ECRR08=2, ECRR09=1, ECRR10=1, ECRR11=6, ECRR12=2, ECRR13=2, ECRR14=2, ECRR15=3, ECRR16=2, ECRR17=5, ECRR18=1, ECRR19=2, ECRR20=5, ECRR21=2, ECRR22=5, ECRR23=2, ECRR24=4, ECRR25=2, ECRR26=6, ECRR27=6, ECRR28=7, ECRR29=7, ECRR30=5, ECRR31=6, ECRR32=2, ECRR33=6, ECRR34=6, ECRR35=6, ECRR36=6.
To compute Chris' anxiety score, you would average the items in the following manner: (ECRR01 + ECRR02 + ECRR03 + ECRR04 + ECRR05 + ECRR06 + ECRR07 + ECRR08 + (8-ECRR09) + ECRR10 + (8-ECRR11) + ECRR12 + ECRR13 + ECRR14 + ECRR15 + ECRR16 + ECRR17 + ECRR18)/18. This will provide an anxiety score of 2.33.
To compute Chris' avoidance score, you would average the items in the following manner: (ECRR19 + (8-ECRR20) + ECRR21 + (8-ECRR22) + ECRR23 + ECRR24 + ECRR25 + (8-ECRR26) + (8-ECRR27) + (8-ECRR28) + (8-ECRR29) + (8-ECRR30) + (8-ECRR31) + ECRR32 + (8-ECRR33) + (8-ECRR34) + (8-ECRR35) + (8-ECRR36))/18. This would lead to an estimated avoidance score of 2.17.
Thus, Chris' estimated anxiety score is 2.33 and his avoidance score is 2.17. This suggests that he is relatively secure.*
This Excel spreadsheet illustrates this process further. In this spreadsheet you can enter various responses to the ECR-R items and see how the computed anxiety and avoidance scores change. (The appropriate responses are automatically reverse keyed behind the scenes in this spreadsheet; you do not need to manually do so.)
If you are familiar with IRT, you can also use the item parameter estimates reported in our 2000 article (and below) to scale people with respect to anxiety and avoidance. This is most easily accomplished by using an existing IRT program for graded response data, such as MULTILOG
* Thank you to Nyssa Lonergan for catching a typo in this equation on June 5, 2013. Item 21 should not be reversed and items 22 and 36 should be.
Q: Do you know where I can find information on the reliabilty and validity of the ECR-R scales?
A. The commonly used estimate of internal consistency reliability tends to be .90 or higher for the two ECR-R scales. As discussed in the 2000 article, IRT analyses suggest that the reliability might be a bit less at the secure end of both dimensions than at the insecure end of the dimensions.
Further information about the reliability and validity of the scales can be found in the following publication(s):
Sibley, C. G., & Liu, J. H. (2004). Short-term temporal stability and factor structure of the revised experiences in close relationships (ECR-R) measure of adult attachment. Personality and Individual Differences, 36, 969-975.If you are interested in IRT-based estimates of the ECR-R items, the alpha (a) and beta (b) values for each item are included in Tables 2 and 3 of the Fraley, Waller, and Brennan article. I have reproduced these tables below for convenience. Note: The alpha value in IRT is a lot like the alpha estimate of reliability in standard test theory. The major difference is that alpha is estimated at each point along the latent continuum. This allows one to see how the reliability of the scores may vary as a function of how secure or insecure an individual is. The graphs of these conditional alphas are provided in the Fraley, Waller, and Brennan (2000) article. The more common alphas tend to exceed .90 for each scale.
for the Experiences in Close Relationships-Revised inventory, and item
response theory parameter estimates.
IRT item parameter estimates for the 18-item ECR-R attachment subscale of Avoidance
Q: Do you have any norms for the ECR-R?
A: There are some ECR-R norms available based on people who have taken the ECR-R on-line. The following statistics are based on a sample of over 17,000 people (73% female) with an average age of 27 (SD = 10). Twenty-one percent of the sample was married.
Here are some of the summary statistics:
Note. The values for age represent the predicted values for variable ages based on a regression model that models avoidance and anxiety as a function of age in years. The equation for avoidance is 2.72 + .008*AGE. The equation for anxiety is 3.67 - .004*AGE.
In the full sample, the correlation between the avoidance and anxiety scales was .40.
Please note that these norms should be taken with a grain of salt. They are based on data that our lab collected online using the ECR-R in the early 2000's. I strongly encourage interested readers to consult Sanjay Srivastava's thoughtful blog entry on the pros and cons of using norms in individual differences research.
Q: Do I need permission to use these scales in my research?
A: No. The scales were published in a scientific journal for use in the public domain. You do not need to contact any of the authors for permission to use these scales in non-commercial research. You may not use the scales for commercial purposes without permission.
Q: The ECR-R items appear to be written for people in romantic relationships. Can I modify the items to make them relevant to other kinds of relationships, such as parental or sibling relationships?
A: You should feel free to modify the items in any way that seems appropriate to you. Many people have modified the items for their research purposes and you should feel free to do the same. We have recently developed a new (and shorter) version of the ECR-R that can be used for a variety of relational targets. To take an online version of that test, please visit this link: ECR-Relationship Structures. Additional information about this measure is available here.
Q: Is there a way to categorize people into a specific attachment category (i.e., secure, fearful, dismissing, preoccupied) on the basis of their scores on the two ECR-R dimensions?
A: In short: There is no "natural" or "correct" way to assign people to attachment categroies or styles.
Over the years I have been interested in the question of whether attachment styles are categorical variables (i.e., matters of kind) or continuous variables (i.e., matters of degree). Taxometric analyses (Meehl, 1995) offer one useful way to address the question of whether a construct is categorical or dimensional. Taxometric analyses on multiple samples and measures--including the strange situation, self-reports of attachment in adults, and the adult attachment interview--suggest that variation in attachment is best modeled with dimensions rather than categories (see Fraley & Waller, 1998; Fraley & Spieker, 2003a, 2003b; Roisman, Fraley, & Belsky, 2007). If you classify people on the basis of their scores, you are necessarily reducing the precision of measurement and lowering your statistical power. Thus, I strongly recommend against classifying people on the basis of their continuous scores.
If you MUST assign people to groups (e.g., your dissertation supervisor is making you do so), there are some good and bad ways of making these assignments. But please keep in mind that, theoretically, there is no RIGHT way to make these assignments because there are no "real" types. The optimal way to make these assignments is probably to do so in a way that ensures that you have an equal number of people in each of your groups. One way to do so is to assign people to groups on the basis of the median (i.e., the point at which half of the cases fall above and half fall below) of each dimension. Thus, you can compute the median score (MAVOID)for avoidance and the median score (MAXN)for anxiety and assign people to the four Bartholomew groups in the following manner:
(a) if the person's anxiety score is < MANX and the person's avoidance score is < MAVOID, then assign him or her to the secure group.
I sometimes find it helpful to classify people for the purposes of explaining the meaning of the scores at the individual level. For example, in one of the examples above, I described Chris as "relatively secure" given his low scores on anxiety and avoidance. However, I strongly recommend against assigning people classifications for the purpose of research. You will best leverage the information you're collecting by treating the variation as continuous and analyzing it as such.
Q: I would like to analyze my data in a continuous fashion. How should I do so?
A: I recommend using basic correlational methods (e.g., correlations, multiple regression, structural equation models).
A standard resesearch question is typically of the form "Do attachment styles predict Y?", where Y is some kind of outcome of interest, such as relationship satisfaction, depressive symptoms, response times in a cognitive task, etc. One common way of analyzing such data is with multiple regression:
Outcome = (constant/intercept) + Beta1*Anxiety + Beta2*Avoidance + (residual variance).
This general analytic framework allows you to study attachment in a continuous manner. Let's assume the Outcome was depressive symptoms and that the estimated values of Beta1 and Beta2 were +.30 and +.20, respectively. These estimates would suggest that people higher in attachment-related anxiety and attachment-related avoidance are more likely to report depressive symptoms. Depending on the theoretical questions you're asking, you might have various expectations regarding whether none, one, or both of the attachment dimensions are necessary to predict the outcome of interest.
Importantly, analyzing the two dimensions simultaenously in a regression framework allows you to interpret your results in a manner that is conceptually aligned with Bartholomew's four attachment prototypes (e.g., secure, fearful, preoccupied, dismissing)--but in a way that allows other regions of the two-dimensional space to be accurately represented. The table below is designed to illustrate the way different patterns of regression coefficients (e.g., beta weights) can be interpreted with respect to various attachment patterns.
Note: It is also possible to test the interaction between attachment-related anxiety and avoidance, although, in my experience, the interaction rarely explains much variance in dependent variables. It is necessary to include the interaction term if you are predicting a pattern of results that cannot be modeled as an additive combination of the two dimensions. For example, if you predict that highly secure people will be high on variable X and that highly dismissing, fearful, and preoccupied people will be low on variable X, it is necessary to include an interaction term to characterize such a pattern because, by definition, this pattern cannot be represented fully as an additive combiation of anxiety and avoidance. For examples of this usage, see Fraley & Bonanno (2004).
Q: The ECR-R appears to be about anxiety and avoidance. What about security? I want to know how secure a person is, not how anxious or avoidant he or she is. Why doesnt the ECR-R assess security?
A: The ECR-R does assess security--don't let the namees of the dimensions mislead you. In Bartholomew's framework (the theoretical model that organizes the ECR-R), security is represented as the "low" ends of these two dimensions. In other words, a prototypical secure individual does not worry about whether his or her partner will abandon him or her (low anxiety) and is comfortable opening up to and depending on others (low avoidance).
If you were to fold the two-dimensions into a one-dimensional space that captured security, the theoretically appropriate way to do so would be to average the anxiety and avoidance scores to tap the dimension that runs at a 45-degree angle across the two-dimensional space. The secure end of this dimension would he low anxiety, low avoidance; the insecure end would be high anxiety, high avoidance.
Q: I have heard that there is an on-line measure based on the ECR-R where students can take the ECR and have their scores on the two dimensions plotted for them.
A: Yes, the site is located at http://www.yourPersonality.net/. The site also provides information on attachment theory more generally, and can be used for educational purposes, or just for fun (or both).
(Important note added Sept 2003: I have recently learned that some people are "scoring" paper-and-pencil versions of the ECR-R by entering their participants' responses into the web page and allowing the web page to score the responses. Please note that the web page automatically randomizes the order of the questions. In other words, each time you load the web page, the items will appear in a different order. If you're simply entering in responses without attending to the order in which the items appear, you will get useless results. Also, sometimes the page only uses some items from the ECR-R; other items are rotated in on an experimental basis. The online test is not meant to be a mechanism by which other researchers can assess attachment for their research purposes. It is an educational tool and one that we use for experimental purposes to collect basic data on item functioning.)
Q: What are the advantages of using the ECR-R over the ECR?
A: We are not sure if there are any advantages at this point. In our article, we were using Item Response Theory to illustrate how theoretical inferences about important developmental issues can vary as a function of the item response properties of the measurement scale--properties that researchers rarely consider. Our analyses convinced me (Fraley) that the attachment scales suffer from some critical problems (i.e., they don't assess "security" with as much precision as "insecurity"), but, ultimately, this is a problem with the item pool available for scale construction. In other words, the ECR and the ECR-R are based on the same item pool, and, consequently, the ECR-R can only improve measurement to a minor degree. We believe that future research must attempt to assess the secure region with more precision. In the meantime, I suspect that the ECR and the ECR-R are, for all practical purposes, identical measures of attachment. We hope people will continue to improve the measurement of adult attachment patterns.
Document last updated: Nov 2012. Updated 'norms' section. July 23, 2005; Dec 10, 2005; Dec 2010. Included some new clafications and the Excel link on Feb 28, 2013. Thanks to Kathy Carnelley for spotting a typo. :-)
Self-Report Measures of Adult Attachment