Abstract
Humans can accurately estimate the likelihood of their perceptual decisions being correct. Here, we asked how well humans can estimate their performance in a duration discrimination task. We investigated whether metacognitive assessment is different for subsecond and suprasecond durations (Lewis & Miall, 2003). On each trial, we presented sequentially two white discs of size 4 dva. The duration of the first disc was constant across trials, and the duration of the second was varied in six different steps. Participants decided which of the two discs was presented for a longer duration. After two consecutive trials, participants indicated which of the two trials they thought their decision was more likely to be correct (confidence forced-choice judgement; Mamassian, 2016). In different blocks, we tested discrimination and confidence inference for three different durations: 200, 500 and 1400 ms. We calculated the sensitivity of the temporal discrimination as the slope of the psychometric function fitted to perceptual decisions. As expected, discrimination sensitivity was lower for longer durations (Gibbon, 1977). Metacognitive performance was estimated by comparing the sensitivity of perceptual decisions that were chosen as more likely to be correct, to the overall sensitivity irrespective of the confidence choices. Sensitivity was reliably better for trials chosen as more likely to be correct. When choosing between two physically identical trials, participants could reliably estimate in which of the two intervals their performance was better. Importantly, confidence gain did not scale with sensory uncertainty. Our results indicate that humans are able to reliably estimate their sensory noise when estimating durations. In addition, metacognitive judgements have different sensitivities for different levels of temporal uncertainty.