Empirical Status of the Cognitive Model 87
ing, and threatening or nonthreatening meaning (e.g., die/dye; weak/week; flu/flew).
Individuals are asked to write down the word they heard presented. In an early study
Mathews, Richards, and Eysenck (1989) found that anxious patients generated signifi-
cantly more threatening spellings than nonanxious patients. This finding has been rep-
licated in other studies (e.g., Mogg, Bradley, Miller, et al., 1994, Experiments 2 and 3).
One could argue that the presentation of ambiguous sentences and other forms
of text comprehension might provide a more accurate representation of the complex
concerns we find in the anxiety disorders than single word stimuli. In these studies
anxiety disorder patients are more likely to generate or endorse threatening rather than
nonthreatening interpretations of the sentences (e.g., Amir, Foa, & Coles, 1998b; D. M.
Clark et al., 1997; Eysenck, Mogg, May, Richards, & Mathews, 1991; Harvey et al.,
1993; Stopa & Clark, 2000; Voncken, Bögels, & de Vries, 2003). On the other hand,
Constans, Penn, Ilen, and Hope (1999) found that non- socially anxious individuals had
a positive interpretation bias for ambiguous social information whereas socially anxious
individuals were more even- handed in their interpretations (see also Hirsch & Mathews,
1997). Brendle and Wenzel (2004) found that socially anxious students had particularly
pronounced negative interpretation bias to self- relevant positive unambiguous passages
and reduced positive interpretation of the same passages after 48 hours. Thus it may be
that both enhanced threat interpretation and reduced postivity bias operate differently,
especially in social phobia, but both are important in characterizing the interpretation
bias in anxiety.
One problem with homophones and ambiguous (or unambiguous) passages is that
the threatening productions of the anxious may reflect a response bias (i.e., tendency to
emit a particular response) rather than an interpretation bias (i.e., tendency to encode
or interpret stimuli in a certain threatening manner; see MacLeod, 1999). Macleod and
Cohen (1993) used a text comprehension task to show that only the high trait- anxious
students had quicker comprehension latency for ambiguous sentences that were fol-
lowed by a threatening continuation sentence. This priming effect indicates that the
high but not the low trait- anxious students were more inclined to impose a threatening
meaning on the ambiguous sentences. A more recent study of homograph pairs (i.e.,
a word with two different meanings; e.g., bank could mean a financial institution or
side of a river) suggests that when threat meanings are primed in generalized social
phobia, this activated interpretative bias may persist longer than it does in nonsocially
anxious individuals (Amir et al., 2005). Furthermore, recent studies employing inter-
pretative bias training suggest a possible causal relation between threat interpretations
and anxiety. Nonanxious individuals trained to make negative or threat interpretations
to ambiguous sentences experienced subsequent increases in state anxiety or anxiety
reactivity (Mathews & Mackintosh, 2000; Salemink, van den Hout, & Kindt, 2007a;
Wilson, MacLeod, Mathews, & Rutherford, 2006). The training effect, however, may
be more pronounced for positive interpretations (e.g., Mathews, Ridgeway, Cook, &
Yiend, 2007; Salemink et al., 2007a), with some studies even finding weak or insig-
nificant effects of negative interpretative training on anxiety levels (Salemink, van den
Hout, & Kindt, 2007b).
In summary there is considerable evidence that the anxiety disorders are character-
ized by a conscious, strategic interpretation bias for threat that is particularly evident
when processing ambiguous information that is relevant to the specific anxiety concerns
of the individual. The fact that this effect has been found in priming studies indicates