Purchase this article with an account.
Brian Maniscalco, Hakwan Lau; On a distinction between detection and discrimination: metacognitive advantage for signal over noise. Journal of Vision 2011;11(11):163. doi: 10.1167/11.11.163.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Informal reports suggest that subjects often find it awkward to rate their confidence after responding “no” in a detection task. This is formally captured by findings that metacognitive sensitivity (i.e., the ability to distinguish correct from incorrect stimulus judgments using confidence ratings) is greater for stimulus presence than for stimulus absence (Fleming & Dolan, 2010; Kanai et al., 2010). However, on a signal detection analysis, such results could be expected to arise from first-order stimulus properties, such as unequal variance between target present and absent distributions, rather than from differences at the level of metacognitive function (Maniscalco & Lau, in review). Here we provide evidence that even when first-order stimulus properties were taken into account by a signal detection model, metacognitive sensitivity was higher for stimulus presence in two “genuine” detection tasks, in which the target absent condition was characterized by an absence of evidence/stimulus energy. A control task in which subjects distinguished between a Gabor pattern and visual noise did not show the same effect– presumably because a transient presentation of visual noise can also be taken as an instance of stimulus presence, thus rendering the task a discrimination between two categories. One interpretation is that a “yes” response in a detection task means that the subject has introspective access to stimulus processing, which can be used to guide accurate metacognition. This account is consistent with the notion that confidence judgments are made by high-level mechanisms that evaluate earlier perceptual processing (Maniscalco & Lau, in review; Rounis et al., 2010; Pleskac & Busemeyer, 2010; Fleming et al., 2010). These results can inform a principled distinction between detection and discrimination: although traditionally both tasks are modeled with a detection model with two distributions, detection is more than just discriminating between two categories (signal vs noise), because the “signal” yields higher metacognitive advantage over “noise.”
This PDF is available to Subscribers Only