This was an exploratory study using the AAI with adults with autism for the first
time. Participants were self-selected, with a higher proportion of women and more people
with mood disorders than might be expected from other research in this area (Barnard et
al., 2001; Ehlers & Gillberg, 1993). As it was not known at the outset whether the
participants would be able to engage with the AAI at all, the sample size was relatively
small, although the analyses had satisfactory power and findings were generally clear-cut.
In all cases, the same patterns of results were replicated across parametric and non-
parametric analyses. The lack of association between measures in the case of the last
research question could be due to limitations in the sensitivity of the theory of mind and
autism measures. Although the ADOS is a well established measure of autistic
symptomatology, its intended use is diagnostic, and the brief duration may not be enough
to pick up all of the relevant information among participants with less obvious difficulties;
a self-report symptom measure could have picked up some more subtle or well-disguised
difficulties. There appeared to be a ceiling effect in participants’ scores on the Strange
Stories, although not on the Eyes task. However, the WASI is a well-established measure
whose validity has been established in people with HFA (Minshew et al., 2005), and like
the ADOS and theory of mind measures, it showed little or no relationship with reflective
function or coherence. This does suggest that IQ may really be unrelated to attachment
status in adults with HFA. There is some evidence among children with autism of a
threshold effect, that is those with IQ below 70 are less likely to be secure but those above
this level are not (Rutgers et al., 2004), while there is only limited evidence for a
continuous relationship where the likelihood of attachment security increases with IQ
(Rogers et al., 1991). This may also hold for adults, that beyond a certain level of
intelligence other factors are more relevant in determining attachment security. The
parallel interview was developed specifically for this study and was not used in the control
23