Confidence-accuracy dissociation via criterion attraction

Humans often have to use different decision criteria in different tasks such as when detecting a mosquito against a white versus a patterned wall. However, it is debated whether people can maintain independent criteria for different tasks. Early work uncovered suboptimal biases when multiple tasks are performed simultaneously, and concluded that in such situations people inadvertently use the same decision criteria across different perceptual tasks. On the other hand, these studies could not measure the criterion location directly and more recent work has questioned whether the same criteria are indeed used across different tasks. To resolve this debate, here we develop a new external noise paradigm that can objectively quantify criterion location across two tasks that optimally require very different criteria. We find strong evidence of “criterion attraction” where the criteria across the two tasks move towards each other but do not become identical. This criterion attraction leads to a large and consistent confidence-accuracy dissociation in the absence of reaction time differences between the tasks. These results unify the seemingly disparate findings in the literature and establish a robust way of inducing dissociations between subjective and objective performance.

Predictive cues reduce but do not eliminate intrinsic response bias

Predictive cues induce large changes in people’s choices by biasing responses towards the expected stimulus category. At the same time, even in the absence of predictive cues, humans often exhibit substantial intrinsic response biases. Despite the ubiquity of both of these biasing effects, it remains unclear how predictive cues interact with intrinsic bias. To understand the nature of this interaction, we examined data across three previous experiments that featured a combination of neutral cues (revealing intrinsic biases) and predictive cues. We found that predictive cues decreased the intrinsic bias to about half of its original size. This result held both when bias was quantified as the criterion location estimated using signal detection theory and as the probability of choosing a particular stimulus category. Our findings demonstrate that predictive cues reduce but do not eliminate intrinsic response bias, testifying to both the malleability and rigidity of intrinsic biases.

Detectability of Multiple Flaws

Two experiments measured detectability, d', and two measures of criterion location for multiple flaws presented separately versus mixed. In the first experiment, the different flaws were the same in type but differed in magnitude; in the second experiment the flaws were of different types. In Experiment 1 there was no difference in d' for either magnitude of flaw as a result of the separate versus mixed conditions, (i.e., each magnitude of flaw had its own d' in both separate and mixed conditions). However, in Experiment 2 all flaws had lower d's in the mixed condition than when each was presented alone. In neither experiment was there any difference in average criterion location between the separate and mixed conditions, nor did individual criterion locations for easy versus hard flaws presented separately differ in Experiment 1. In Experiment 2 there was a difference in individual criterion locations for the different flaws presented separately, such that for the hardest to detect flaw, false rejection error rate was lower and miss rate higher than for either of the two flaws with larger d's.

Development of a Mechanism for Detection of Error and Retention of a Graded Motor Response

Several methods for developing an internal mechanism for error detection were compared to determine the extent to which a self-paced response could be acquired and retained without knowledge of results. During training, blindfolded subjects attempted to learn a positioning response under one of three conditions. One group was presented with the criterion end location by means of a stop and then required to recall the location with the stop removed. A second group moved the total distance of the track and then was instructed to replace the slide in a location representing a specific fraction of the total. A final group of subjects learned the criterion location by using post-response knowledge of results administered on each trial. Following acquisition subjects were given retention trials involving only repetition of the acquired response. The development of a strong mechanism for error detection was indicated only for those groups presented with the standard or with post-response knowledge of results. During retention trials (without knowledge of results), performance of the group required to learn the criterion by estimating a fraction of the total range was characterized by significantly greater response biasing and less consistency than the other conditions. The findings were discussed in terms of several predictions from Adams' (1971) closed-loop theory of motor learning.