Collins & Read (1990) built on the earlier work of Hazan & Shaver (1987) and Levy & Davis (1988) when they developed a measure which they called the Adult Attachment Scale (AAS). A revised version of the measure was then used in a further study by Collins (1996).

This scale was developed by decomposing the original three prototypical descriptions (HazanShaver‚ 1987) into a series of 21 items. Participants scored each of these items according to how ch‎aracteristic it was of themselves‚ using a five item Likert-type scale with values ranging from "not at all" to "very".  The 21 items that comprise the original version of the measure (Collins & Read‚ 1990)‚ followed by a code indicating the Hazan and Shaver descriptions from which they were taken‚ are as follows: &

Notes:

Factor analysis of the results led to the emergence of three main factors which were interpreted by the authors as capacity to be close (close); capacity to depend on others (depend); and anxiety over relationships (anxiety).  Of these dimensions‚ the close and depend dimensions correlated fairly closely (r = 0.41‚ N = 406).  Examination of the questionnaire components that had contributed to each of these factors indicated that the depend and close factors contained items drawn from the secure and avoidant attachment style descriptions whilst the anxious factor contained items drawn from the secure and anxious/ambivalent descriptions.

Collins and Read undertook an extensive statistical analysis of the data they gathered‚ in order to investigate the relative merits of viewing attachment style as dimensions or nominal categories.  To investigate the dimensional perspective‚ they used a discriminant analysis technique and identified two significant discriminant functions. The first distinguished the avoidant from the secure and anxious styles and the second distinguished the anxious from the secure and avoidant styles.  Using this technique to reclassify subjects based on questionnaire item scores resulted in 73% of the sample being correctly classified.

To check for descrete classifications‚ they used cluster analysis.  A technique to determine the optimum number of clusters indicated that either three or four clusters should be used. Collins and Read opted for a three cluster solution and found the ch‎aracteristics of the resulting clusters corresponded well with Hazan and Shaver's three categories of attachment.

Even though their three dimensional factors (close‚ depend and anxiety) did not directly correspond well to any existing categorical or dimensional models of attachment theory‚ Collins and Read concluded that “we believe the dimensions measured by the Adult Attachment Scale capture much of the core structures that are thought to underly differences in attachment styles.” (p. 650).

Note that later researchers have criticised Collins and Read's three dimensions on the grounds that they do not conform to either the three- or the four-style classification systems.  Only the anxiety dimension was as expected (corresponding to anxious/ambivalent) whilst the close and depend dimensions (which were correlated) apparently represented two different measures of avoidance (Carver‚ 1997)

Reported reliability scores from other studies using this instrument have been moderate (Chongruska1994; Sperling‚ Foelsch & Grace‚ 1996) and the latter authors also found convergent validity between this measure and their own Attachment Style Inventory (SperlingBerman‚ 1991). & ‚