Abstract
The internal representation of stimuli is imperfect and subject to biases. Noise is introduced at initial encoding and during maintenance, degrading the precision of stimulus estimation. Stimulus estimation is also biased away from recently encountered stimuli (adaptation). According to a Bayesian framework, one should expect bias to be greater when stimulus representation (and thus precision) is poor. We tested for this effect in individual difference measures. 101 subjects performed an on-line experiment (https://cfn.upenn.edu/iadapt). During separate face and color blocks, subjects performed three different tasks: an immediate stimulus-match (15 trials), a 5s delayed match (30 trials), and a 100ms delayed match following 5s of adaptation (30 trials). The stimulus space was circular, and subjects entered their responses using a color/face wheel. Bias and precision of responses, while accounting for random guesses, were extracted by fitting a mixture of von Mises distributions. Two blocks of each measure were obtained, allowing for tests of measure reliability. 33 subjects were excluded due to poor model fit. In the remaining subjects, average precision across all tasks was found to vary significantly between subjects (p< 0.001), but not vary as a function of task or material. Precision was a reliable subject property (r=0.67 over two measures). The adaptation manipulation induced the expected bias in responses (mean±SEM subjects: 7.7±0.5° colors, 4.2±0.6° faces). The magnitude of this bias varied between subjects, and this too had good re-test reliability (color r=0.69; faces r=0.52). Across subjects, there was a negative correlation between mean precision and bias (r=-0.29, p=0.016). While a significant correlation between face and color bias was found (r=0.40,p< 0.001), this did not account for the entire reliable variance of individual differences for the two materials. At the level of individual differences, the precision of perceptual representations negatively predicts the magnitude of adaptation bias.
Meeting abstract presented at VSS 2015