The Problem with Self-Report Attachment Measures.

Richard J. Atkins - 28th August, 2006.

Some authors have criticised attachment measures, notably the RSQ, ASM, and ASP measures, as being based on conventional rather than empirical categories of attachment.  This note sets out my views on this and outlines what I see as a common problem in the development of psychometric measures.

Attachment measure development across the late 80's and 90's was largely driven by study after study using factor analytic methods to identify dimensions of attachment based on responses to a pool of questions.  In many cases, selection of the initial pool of items for each study was based on those that had shown strong factor loadings in previous studies, and the resulting measure retained those items that had strong factor loadings in the present study.  While, on the face of it, this seems an appropriate way to go about developing a measure it has one potentially fatal flaw.  Repetitively selecting items based on factor analysis will empirically identify those items that show the strongest links with each other based on the main constructs being measured, irrespective of what those constructs actually are.  If current relationship functioning is the main influence on individual's responses to the items, the dimensions identified and the items selected will be those that measure this construct, irrespective of the construct you set out to measure. Across repeated studies such measures will converge to a list of items that has high reliability and is highly specific to this construct. In the case of attachment there have been notable attempts to introduce items derived from Bowlby's original conception of attachment (e.g. items measuring secure base and safe haven functioning in romantic relationships; Carver, 1997) which have not been incorporated within the empirically derived dimensions because, despite being fave-valid attachment items, they do not fit with the existing items and dimensions derived from prior empirical work.

In further support for this, there is ample evidence that self-report attachment measures show stronger relationship with variables tapping current relationship functioning (e.g. marital satisfaction) than with reports of early caregiver relationships (e.g. maternal provision of a secure base), while for non-self report measures such as the AAI this situation is reversed.  Overall, I agree with the concerns expressed by Feeney & Noller (1996) that self-report measures may be a better measure of state relationship functioning rather than trait-like characteristic and I believe this has been due to the extensive use of empirical methods that have selected precisely those items that have been influenced by the stronger, proximal influence of current relationships rather than the more distant influence of early relationships and any consequent traits.  This being the case I see no harm and possibly a number of advantages in using an attachment measure that has been constructed to be a face-valid measure of attachment as conceptualised by Bowlby, rather than an empirically developed measure that may be measuring current relationship functioning interpreted within an attachment framework, rather than a measusuring underlying attachment characteristics themselves.