group, so there are some questions about how it would be expected to relate to AAI
responses in a sample without autism. For example, one concern is that it was shorter than
the AAI, which could account for the difference in coherence scores. Ideally, it would
have been preferable to have all the same measures for the controls as for the participants,
so that the patterns of relationships in discourse style, IQ and so on could be directly
compared across both groups.
Future research directions
It was suggested in the Introduction that children with autism who form secure
attachments could be doing so in different ways from other children. The same may yet
hold true of autistic adults with secure attachments, although the present study found no
evidence of this. A study with more participants might be able to identify a sufficiently
large group of secure adults with autism to allow group comparisons, rather than needing
to rely on continuous measures as proxies for attachment security as this study has done.
It was predicted that the AAI would be a particularly difficult task for adults with autism
because of its heavy reliance on discourse coherence, mentalising and conversational
pragmatics. There are now semi-projective measures of attachment in adults (e.g. George
& West, 2001) that rely less strongly on narrative than the AAI. If adults with autism
demonstrated the same level of attachment security on these measures as found in the
current study, it would give a measure of convergent validity and demonstrate that they
were really being scored according to their states of mind with respect to attachment,
rather than on how difficult they found the AAI. Using the AAI among adults with autism
for the first time, this study has succeeded in demonstrating that the participants were able
to respond to the interview in a way that could be classified, that some of them could be
classified as securely attached and that their responses were at least partly independent of
24