A Bluffer's Guide to using Self-Report Adult
Attachment Measures in your Research Project
Richard J. Atkins
Originally Published in 1999, last update in 2003
When I was a lecturer in the psychology department at Thames Valley University (TVU), each year a few students would want to undertake coursework and dissertation research projects using adult attachment style as one of the variables. Most of these students originally planned to use a measure they had encountered in general psychology textbooks, typically the original, twenty-year old three-item forced-choice measure (Hazan & Shaver, 1987). Most of these students appeared to be unaware of the problems associated with using a categorical measure or of the range of other options available. This led me to write a paper for the TVU web site laying out the history of self-report attachment measures and giving some advice on which measures to use. This paper is largely based on the original written for TVU, but has been cut-down, revised and integrated into the Attachment Theory Website framework.
Please note that only self-report, questionnaire based attachment measures are considered. These are used to determine the attachment style of adult or adolescent participants, by questioning the participant on their experience and attitudes towards adult (usually romantic) relationships. This type of measure comes from the social/personality psychology approach to attachment which tends to be interested in conscious attitudes that can be tapped using self-report measures. Furthermore, this class of instruments tend to be based on attitudes to adult, peer or romantic attachments, rather than parent-child and early attachment experiences. Other adult measures exist, particularly the Adult Attachment interview (George, Kaplan & Main, 1985; 1996) that use clinical-type interviews to tap participants subconscious working models . These are not suitable for use by undergraduate researchers, who are not generally trained in clinical interview techniques, so they are not covered here.
List of Measures Considered
The list of measures that I have examined in producing this is as follows:
- 10 Item Attachment Scales (10AS; Brennan, Shaver & Hazan, 1989; Brennan & Shaver, 1995)
- Adult Attachment Questionnaire (AAQ; Hazan & Shaver, 1987; 1990)
- Adult Attachment Scale (AAS; Collins & Read, 1990; Collins, 1996)
- Attachment Style Measure (ASM; Simpson, 1990)
- Attachment Style Prototypes (ASP; Levy & Davis, 1988)
- Attachment Style Scales (ASS; Becker & Billings, 1997)
- Experience in Close Relationships Scale (ECR; Brennan, Clark & Shaver, 1998)
- Experience in Close Relationships Scale - Revised (ECR-R; Fraley, Waller & Brennan, 2000)
- Measure of Attachment Qualities (MAQ; Carver, 1997)
- Relationship Questionnaire (RQ; Bartholomew & Horowitz, 1991)
- Relationship Scales Questionnaire (RSQ; Griffin & Bartholomew, 1994)
N.B. Click on each measure's abbreviation for a page specifically on that measure.
So Which Measure Do I Use?
The first question is probably whether to use a categorical or a dimensional measure.
Categorical measures are easy and fast to administer, which can appear a distinct advantage in undergraduate projects where obtaining large sample groups can be difficult. This, however, can be a false economy. When a typological measure is used to measure a continuous variable this introduces errors and also problems with statistical such as:
- A lot of the available detail in the raw data is lost and the result of this is lower statistical power and less chance of getting a statistically significant result unless using a larger sample. For example, think about the range of outside temperatures over the year and how much detail you lose if every day is simply classified as 'hot' or 'cold'.
- The boundaries created by imposing a typology are arbitrary and may be different for subtly different measures. This means different measures that may be tapping the same constructs may have different arbitrary boundaries so an individual who is 'secure' (but only just) on one measure may end up classified as 'anxious' on another and 'avoidant' on yet another measure.
- Misclassification errors can arise within a measure where you have participants whose continuous scores would place them near the boundaries of two attachment categories. Such participants would be considered marginal using continuous scores and would not be influential in any statistical results. Using a categorical measure, marginal participants must end up in one classification or another, may end up in the wrong category, and will be treated as no different to someone who is an extreme example of that particular classification.
- Given the distribution into different attachment categories is not even (e.g. using the AAQ we get about 60% secure, 25% anxious and 15% avoidant), a large sample size will be needed to get a reasonable number of people in the rarer categories (e.g. avoidant) to allow worthwhile comparisons with the more common categories. For example, if we want to look at sex differences in attachment style with at least 20 people in every analysis group, using a continuous scale such as the ECR we would need 40 people (2 independent groups of 20 and they all give a score on anxiety and avoidance) but using the categorical AAQ we would need 268 participants to have an average chance of getting 20 males and 20 females who are classified as avoidant.
As regards the question of whether attachment should be conceptualised as categorical or continuous, many authors have argued that attachment is continuous and should be measured using dimensions scores rather than classifications (Brennan, Shaver & Tobey, 1991; Collins & Read, 1990; Brennan, Clark & Shaver, 1998). Apart from argument, Fraley & Waller (1998) have provided strong empirical support for the idea that attachment traits are evenly (normally) distributed across two continuous dimensions.
If you do decide to go ahead and use a categorical measurement despite all this, then the choice is probably between the two most widely used forced-choice measures: the AAQ if you want a three-category measures and the RQ if you want four categories of attachment. Of these, the AAQ is probably the single most widely used attachment theory measure. Much of the published research to date has used this measure which allows direct comparison between your research and that of other authors. Whilst the RQ has proven a less popular measure in the literature, it has the advantage of having a stronger relationship with the two theoretical dimensions of avoidance and anxiety which are believed to underly attachment style (Bartholomew, 1990; Bartholomew & Horowitz, 1991). Whilst the three categories of the AAQ very approximately map onto the RQ (Brennan, Shaver & Tobey, 1991), RQ adds a further category of avoidance that provides about the only chance of finding a gender difference in attachment categories (but see my comments above about minimum sample sizes).
Of the dimensional measures, many have suffered from less-than-ideal internal reliability (ASM, MAQ), from being based on conventional rather than empirical factors (RSQ, ASM, ASP; although see this note on why I don't think this is a bad thing), have a large number of individual items (10AS), or have simply been superceded by later research (AAS, ASS). Once all these factors are taken into account, the only remaining measure is the Experience in Close Relationships Scale (ECR or the more recently revised version of the same scale ECR-R. Both of these scales benefit from being relatively short (32 items) and having high internal reliability scores on both of the dimensions measured. The authors originally presented the ECR as "an all-purpose reply to future attachment researchers who wish to use self-report measures" (Brennan, Clark & Shaver, 1998, p. 46). The only disadvantage of this measure that I can see is that it only provides two scores on the global constructs of avoidance and anxiety. Just like personality, attachment can be conceptualised in terms of both traits and facets. If you want to investigate the facets of attachment (if, for example, you want to differentiate between proximity seeking and 'clinginess'), it may be worth looking at one of the multi-dimensional measures such as the 10AS. Note, however, that a more detailed measure means more separate items and consequently a higher number of participants to ensure adequate statistical power.
Other technical reviews of attachment measures are provided by Garbarino (1998), Lopez & Gover (1993), and Lyddon, Bradford & Nelson (1993).
Some of the authors above provide further information on their own and other measures on the web. Check out the pages provided by Shaver & Fraley (unknown) and Shaver & Fraley (unknown).