Examining the influence of cue saliency in multiple-cue probability learning paradigms, we used a modified weather prediction task (Knowlton et al., 1994) to conduct two experiments (Ntotal = 258) on (in)congruency-dependent differential learning dynamics. In Experiment 1, the subjects had four faces as cues to predict a binary outcome ("truth" vs. "lie"). The hierarchy of a face’s trustworthiness was either congruent or incongruent to the face’s forecasting validity for the positively valenced outcome “truth”. Consequently, people experiencing incongruency learned the ecological validities of trustworthy faces being associated with the outcome "lie" and untrustworthy faces associated with the outcome "truth". In contrast, people experiencing congruency learned the associations vice versa.
Notably, the incongruency group yielded lower accuracy rates than the congruency group, both groups displaying unique learning trajectories. As evidenced by distinct cue utilization curves detected by the rolling regression method of Lagnado and colleagues (2006), incongruency causes more uncertainty and a delayed strategy optimization. In Experiment 2, using the same experimental design as in Experiment 1 except for introducing two different scenarios (witness- and gambling door-scenario) with different binary outcomes ("truth" vs. "lie" and “win” vs. “loss”), we explored the true nature of the effect. Specifically, we scrutinized whether it reflects a socially relevant dimension affecting a social judgment criterion, or whether it is merely based on the association of similarly valenced cues/outcomes (positive vs. negative).