Integrating explicit reliability for optimal choices: effect of trustworthiness on decisions and metadecisions

In everyday life, decision-makers need to integrate information from various sources which differ in how reliable the correct information is provided. The present study addressed whether people can optimally use explicit information about the trustworthiness of the information source and whether people know how information sources affect their decisions. In each trial, we presented a sequence of six information sources which indicated a correct colour (red or blue) each with a different level of reliability. Each source was explicitly labelled the reliability percentage, i.e., the likelihood that a source provides correct information. Participants were asked to decide which colour was more likely to be correct. In the first series of three experiments, we found that participants failed to make use of explicit reliability cues optimally. In particular, participants were less able to use reliably wrong information sources (reliability below 50%) than reliably correct information sources (reliability above 50%), even though these two types of information sources were equally informative. In addition, participants failed to ignore the colour displayed by unreliable sources (50% reliability), although these sources gave just random information for a binary decision. In the second series of two experiments, we asked participants, after each choice, to report their subjective feelings of whether they followed or opposed a colour suggested by sources with a specific reliability percentage. We found that the ratings of the participants’ influence report tracked the amount of evidence which supported the choice they just made. Further, participants were able to introspect their own choice bias guided by unreliable information sources. These findings suggest that human choice behaviour deviates from Bayesian integration. However, people have a good metacognitive monitoring of how their decisions are driven by external stimuli.