Abstract
Confidence judgment involves subjective evaluation of uncertainty in performing a task. For a task that requires the visual system to integrate multiple elements in the stimulus, there are two sources of uncertainty: Noise pertaining to local, individual elements affects lower levels of processing, while variability across elements affects the global integration of elements at higher levels of processing. How does noise at different processing levels affect confidence judgments? The present study addresses this question using a local-global-noise paradigm. The stimulus was a multiple-aperture motion pattern, consisting of an array of randomly-oriented, drifting-Gabor elements, superimposed with dynamic pixel noise. Global signal-to-noise ratio (GSNR) was defined as the ratio between the coherently-moving (signal) elements and randomly-moving (noise) elements. Local signal-to-noise ratio (LSNR) was defined by the contrast ratio between Gabor and noise pixels. In each trial, observer judged the global rotational direction (clockwise or counterclockwise), and simultaneously gave a confidence rating (high or low) on the direction judgment. Four levels of LSNR were fixed for each observer based on a prior calibration session. At each LSNR level, GSNR threshold was varied using an adaptive-staircase procedure, so that direction-judgment accuracy converged to 75%. For a constant direction-judgment accuracy, a tradeoff between local and global noise was found, suggesting that perceptual sensitivity can be equated between a low-LSNR-high-GSNR stimulus and a high-LSNR-low-GSNR stimulus. Analysis on confidence judgments was based on the signal-detection theoretic approach (Maniscalco & Lau, 2012). No systematic difference in meta-d' was found between low-LSNR-high-GSNR and high-LSNR-low-GSNR stimuli. However, confidence criterion tended to be more liberal for high-LSNR-low-GSNR stimuli than for low-LSNR-high-GSNR stimuli, suggesting that observers tended to be more confident when global uncertainty was high than when local uncertainty was high, despite similar first-order performance. These findings suggest the level at which uncertainty exists affects the confidence criterion, but not metacognitive sensitivity.
Meeting abstract presented at VSS 2017