Near-optimal metacognition across the visual periphery

How effectively does perceptual confidence track task accuracy across the visual field? Previous research has provided an array of findings, including underconfidence in the visual periphery (Toscani, Mamassian, and Valsecchi, 2021), overconfidence in incorrect trials (Odegaard, Chang, Lau, and Cheung, 2018), and anecdotal accounts of being surprised at how poorly we perform visual tasks at eccentric locations (Cohen, Dennett, and Kanwisher, 2016). In this investigation, we aimed to systematically characterize metacognitive efficiency for visual detection judgments up to forty degrees away from fixation. On each trial, twenty-four participants performed a 2AFC task to report signal (a Gabor) or noise at a specific location along a horizontal axis, and rate confidence in their judgment. On most trials, participants were cued to the location of the upcoming stimulus (valid trials), but on a small proportion of trials, participants were cued to an incorrect location (invalid trials). Results showed that, as expected, perceptual sensitivity declined with eccentricity, and valid trials had higher perceptual sensitivity than invalid trials. Importantly, metacognitive efficiency, as measured by the metric meta-d’/d’, was nearly optimal at each eccentric location, showing that confidence judgments effectively tracked task accuracy. Additionally, we revealed a unique profile of decision criteria across eccentric locations as a “check-mark” pattern emerged, with very conservative criteria at the 40-degree location, and unbiased criteria closer to fixation. Decision criteria were minimally impacted by attentional cueing, and a second experiment without cues replicated the “check-mark” pattern across eccentric locations. Finally, results showed that declining metacognitive sensitivity in eccentric locations was driven primarily by declining confidence in correct trials, and surprisingly consistent confidence on incorrect trials across eccentricities. Together, these results challenge accounts of liberal criteria and suboptimal confidence in the periphery, and reveal that our sense of confidence may track performance decrements more effectively than previously supposed.