The role of event uncertainty on inferences predictive of threat was investigated in high and low trait anxious individuals. Participants read context sentences predicting threat or nonthreat outcomes. They subsequently named target words that were consistent with or unrelated to prediction. In Experiment 1, with predictability relatively low, anxious participants showed clear threat bias in their inferences: Although nonthreat targets were unaffected by context, shorter naming latencies were found for threat target words that followed a threat predicting context. A low anxiety group showed an opposite effect, that is, facilitation only for nonthreat words, suggesting an avoidance (of threat) bias. In Experiment 2, under higher predictability, this bias disappeared, as both high and low anxious groups performed similarly. The relevance of these data for different models of selective processing in anxiety is discussed. Of particular pertinence is the finding that, with increasing stimulus threat, low anxious participants no longer show avoidance; instead, they infer threat in a way similar to the high anxious. This suggests that the difference between the high and the low anxious persons resides in the threshold at which stimulus threat input is processed.
