Decisions are based on less information than metacognitive judgments in multialternative contexts

Abstract

Humans often face decisions between multiple alternatives. In these contexts, some evidence suggests that only the alternative with the highest evidence is represented by the decision system. However, other findings indicate that unchosen alternatives’ information remains available for decision computations. To evaluate how much information from unchosen alternatives is accessible by the decision system, we employed a second-guess paradigm: When participants selected an incorrect alternative, they were given a second opportunity to make a new choice. By fitting computational models to data from two preregistered experiments involving four (Experiment 1) and 12 (Experiment 2) alternatives, we found evidence for an intermediate position: After the first decision is made, noise corrupts the evidence from the initially unchosen options, suggesting that the decision system cannot access all the sensory evidence available to perform a second decision. We extended this finding by fitting the models to two previously published data sets involving different stimuli and numbers of alternatives (six and three) and found concordant evidence. In addition, we also evaluated the amount of information accessible by the metacognitive system, responsible for monitoring our behavior and reflecting upon the correctness of our decisions. We found that incorporating a separate channel of evidence unaffected by noise for metacognitive computations improves model fitting, suggesting that the decision system accesses less evidence than the metacognitive system. These results reconcile previous conflicting findings in multialternative decisions and highlight a dissociation between decision making and metacognition, offering new insights into the fundamental constraints of decision processes and the relative robustness of metacognitive evaluations.