In a dot-probe task, two cues-one emotional and one neutral-are followed by a probe in one of their locations. Faster responses to probes co-located with the emotional stimulus are taken as evidence of attentional bias. Several studies indicate that such attentional bias measures have poor reliability, even though ERP studies show that people reliably attend to the emotional stimulus. This inconsistency might arise because the emotional stimulus captures attention briefly (as indicated by ERP), but cues appear for long enough that attention can be redistributed before the probe onset, causing RT measures of bias to vary across trials. We tested this hypothesis by manipulating SOA (stimulus onset asynchrony between onset of the cues and onset of the probe) in a dot-probe task using angry and neutral faces. Across three experiments, the internal reliability of behavioural biases was significantly greater than zero when probes followed faces by 100 ms, but not when the SOA was 300, 500, or 900 ms. Thus, the initial capture of attention shows some level of consistency, but this diminishes quickly. Even at the shortest SOA internal reliability estimates were poor, and not sufficient to justify the use of the task as an index of individual differences in attentional bias.
Keywords: Attention; Attentional bias; Dot probe; Emotion; Reliability; Threat bias.