The time course of the bias in predictive inferencing as a function of
anxiety was examined. Ambiguous sentences (concerned with ego-threat,
physical-threat, or nonthreat events) were presented, followed by dis
ambiguating sentences in which a target word either confirmed or disco
nfirmed the consequences implied by the ambiguous context. High- and l
ow-anxiety subjects read the sentences at their own pace with the movi
ng-window procedure. Effects on word reading times at different points
were measured. Priming effects occurred for high-anxiety subjects whe
n reading disambiguating ego-threat-related sentences: There was great
er relative facilitation for confirming than for disconfirming ego-thr
eat versions in high-anxiety subjects, compared with physical-threat a
nd nonthreat versions, and with low-anxiety subjects. Because these ef
fects were not observed in the target word itself but in the post-targ
et region and the last word of the disambiguating sentence, we conclud
ed that the bias towards ego-threat predictive inferences does not occ
ur during the initial processing of the ambiguous information, but rat
her, is a post lexical delayed phenomenon. The mechanisms of such bias
are discussed in relation to current models of inferences in reading.